Control method and vehicle

Abstract

The control method and the vehicle provided by the application are related to the technical field of vehicle safety control. The method comprises the following steps: in response to a target object entering an induction space of a projected capacitive sensor array of a target device, a corresponding capacitive image sequence is acquired, the capacitive image sequence comprises a plurality of images of changes of capacitive values of each coordinate point on the projected capacitive sensor array caused by the target object over time; from the capacitive image sequence, a multi-dimensional feature is analyzed, and the safety hidden danger of target misclassification caused by single-dimensional perception deficiency is solved; and based on the multi-dimensional feature, a control strategy of the target device is generated, sufficient and reliable input information is provided for generation of the safety control strategy, a technical difficulty that an effective safety control strategy cannot be realized due to insufficient perception information in the prior art is solved, and the reliability of target object discrimination and the safety of target device control in a complex scene are improved.

Description

Technical Field

[0001] This application relates to the field of vehicle safety control technology, and more specifically, to a control method and a vehicle. Background Technology

[0002] Safety control of target devices relies on accurate perception of the target objects inside them. In the vehicle field, the Occupant Classification System (OCS) is a key component of modern automotive passive safety systems. Its main function is to automatically identify the type of target object on the vehicle seat (such as adults, children, objects, etc.) and provide this information to the control unit to generate differentiated control strategies aimed at achieving optimal protection.

[0003] Currently, existing occupant classification systems generally adopt seat-embedded sensing technology, which involves embedding a pressure sensor matrix or strain gauge inside the seat cushion to collect weight data of the target object to distinguish the type of the target object.

[0004] However, existing solutions rely solely on single-dimensional information such as the target object's weight or pressure distribution, making them unable to make reliable judgments in complex scenarios. They struggle to distinguish between target objects of similar weight but different types (e.g., a child and a heavy object), or target objects of the same size but different types (e.g., a real person and a humanoid doll or robot), and cannot perceive the target object's crucial spatial posture (e.g., whether it exhibits vital signs, whether it is leaning forward or sitting sideways). This lack of perceptual dimensions not only creates safety hazards due to misclassification but also renders safety control strategies ineffective due to a lack of necessary input information. Summary of the Invention

[0005] This application provides a control method and vehicle that solves the safety hazard of target misclassification caused by insufficient single-dimensional perception. It generates control strategies for target devices based on multi-dimensional features, providing sufficient and reliable input information for the generation of safety control strategies. This solves the technical problem that existing technologies cannot achieve effective safety control strategies due to insufficient perception information, and improves the reliability of target object identification and the safety of target device control in complex scenarios.

[0006] In a first aspect, a control method is provided, comprising: in response to a target object entering the sensing space of a projected capacitive sensor array of a target device, acquiring a corresponding capacitance image sequence, the capacitance image sequence including multiple images showing the change of capacitance values ​​at various coordinate points on the projected capacitive sensor array over time caused by the target object; parsing multi-dimensional features from the capacitance image sequence, the multi-dimensional features being features parsed from the temporal and spatial domains of the capacitance image sequence; and generating a control strategy for the target device based on the multi-dimensional features.

[0007] In the above technical solution, by responding to the sensing space of the projected capacitive sensor array of the target device when the target object enters, a capacitance image sequence is obtained, consisting of multiple frames of images showing the capacitance values ​​of each coordinate point of the sensor array changing over time due to the target object. This abandons the inherent mode of existing solutions that rely solely on single-dimensional information such as weight data and pressure distribution for perception, expanding the dimensions of information acquisition from the source of perception. Furthermore, multi-dimensional features are parsed from the temporal and spatial domains of this capacitance image sequence. With the help of these spatiotemporal dual-dimensional feature information, it is possible to distinguish target objects with similar weight but different types, or with the same body shape but different categories. It is also possible to perceive the vital signs of the target object, key spatial postures such as leaning forward or sitting sideways, solving the safety hazard of target misclassification caused by the lack of single-dimensional perception. Finally, a control strategy for the target device is generated based on these multi-dimensional features, providing sufficient and reliable input information for the generation of safety control strategies. This solves the technical problem that existing technologies cannot achieve effective safety control strategies due to insufficient perception information, improving the reliability of target object identification and the safety of target device control in complex scenarios.

[0008] In conjunction with the first aspect, in some possible implementations, multi-dimensional features are extracted from the capacitance image sequence, including: processing the capacitance image sequence to determine the projection area of ​​the target object on the projected capacitance sensor array; and performing temporal and spatial domain analysis on the capacitance image sequence corresponding to the projection area to obtain multi-dimensional features.

[0009] In the above technical solution, by performing preprocessing operations such as noise reduction, differential calculation, and connected component filtering on the capacitance image sequence, environmental interference and invalid noise signals are effectively eliminated, and the projection area corresponding to the target object on the projected capacitance sensor array is determined. This enables the positioning and boundary division of the effective sensing range of the target object, avoiding interference from non-target area data in subsequent analysis. On this basis, further temporal change analysis in the time domain and spatial distribution analysis in the spatial domain are performed on the capacitance image sequence corresponding to the projection area. This allows for the extraction of multi-dimensional features, including the target object's position, shape, dynamic changes, and coupling strength. This ensures the relevance of multi-dimensional feature extraction and improves the reliability of multi-dimensional feature data, laying a data foundation for the subsequent generation of scientific and adaptive target device control strategies.

[0010] In conjunction with the first aspect, in some possible implementations, the capacitance image sequence is processed to determine the projection region of the target object on the projected capacitance sensor array, including: thresholding the capacitance value of each coordinate point in the capacitance image sequence based on a preset capacitance threshold to obtain a binary image sequence; analyzing each frame of the binary image sequence to aggregate adjacent pixels with the same pixel value in each frame into a connected region; and extracting the connected region with the largest area from each connected region as the projection region of the target object.

[0011] In the above technical solution, thresholding is performed on the capacitance values ​​of each coordinate point in the capacitance image sequence based on a preset capacitance threshold. This effectively removes environmental interference and minor noise signals, transforming the original capacitance data into a binary image sequence with clear boundaries and prominent features, thus achieving effective separation of the target sensing signal and the background signal. Furthermore, connected component analysis is performed on each frame of the binary image sequence, aggregating adjacent pixels with the same pixel value into connected regions. This allows for the localization of areas where capacitance changes caused by the target object are concentrated, avoiding recognition bias caused by discrete points. Finally, the connected region with the largest area is extracted from each connected region as the projection region of the target object. This eliminates the influence of stray interference regions, ensuring that the extracted projection region accurately reflects the actual sensing distribution of the target object. This provides a stable and reliable regional basis for subsequent feature extraction and intent recognition, improving the accuracy of subsequent feature analysis and control decisions.

[0012] In conjunction with the first aspect, in some possible implementations, the capacitance image sequence corresponding to the projection area is analyzed in both the temporal and spatial domains to obtain multi-dimensional features, including: obtaining liveness features based on the temporal changes of the capacitance image sequence corresponding to the projection area; and calculating spatial contour features based on the spatial distribution of capacitance values ​​in a single frame capacitance image of the projection area. The multi-dimensional features include: liveness features and spatial contour features.

[0013] In the above technical solution, based on the temporal changes of the capacitance image sequence corresponding to the projection area, the liveness features used to distinguish between real live objects and non-live interference objects can be analyzed, thereby eliminating interference factors such as spurs and static foreign objects from the temporal dimension and improving the realism and anti-interference ability of target object detection; at the same time, based on the spatial distribution of capacitance values ​​in a single frame capacitance image of the projection area, the spatial contour features characterizing the shape, size and spatial distribution of the target object can be calculated, providing a reliable static quantitative basis for the shape determination and regional positioning of the target object; the multi-dimensional features in this application can be composed of liveness features and spatial contour features, and through the organic combination of temporal dynamic features and spatial static features, a multi-dimensional and three-dimensional feature characterization of the target object can be achieved.

[0014] In conjunction with the first aspect, in some possible implementations, the multi-dimensional features include: liveness features, which are signal features of the periodic physiological activities of a living organism; and liveness features are obtained by analyzing the temporal changes of the capacitance image sequence corresponding to the projection region, including: extracting signals from the capacitance image sequence corresponding to the projection region to obtain the temporal signal of the capacitance change in the projection region; performing spectral analysis on the temporal signal to extract the periodic fluctuation features of the temporal signal within the preset vital sign frequency band; and using the periodic fluctuation features as liveness features.

[0015] In the above technical solution, the multi-dimensional features include living features, which are signal features characterizing the periodic physiological activities of living organisms. In the feature extraction process, the capacitance image sequence corresponding to the projection area is first analyzed and extracted to obtain a time-domain signal that reflects the capacitance change pattern within the projection area. Then, the time-domain signal is subjected to spectrum analysis to extract the periodic fluctuation features of the time-domain signal within the preset vital sign frequency band. These periodic fluctuation features are directly used as living features. Through this feature extraction method based on capacitance time-series changes and spectrum analysis, the unique physiological activity signals of living organisms can be captured, and living targets can be reliably distinguished from non-living interference objects. This avoids the problem of misidentification caused by non-living objects triggering sensor signals, while improving the stability and recognizability of the extracted living features.

[0016] In conjunction with the first aspect, in some possible implementations, the spatial contour features include: geometric features, which are the shape, size, and positional distribution characteristics of the projected region of the target object on the projected capacitive sensor array; the spatial contour features are calculated based on the spatial distribution of capacitance values ​​of the projected region in a single-frame capacitive image, including: calculating the area of ​​the projected region and the coordinates of the centroid of the projected region on the projected capacitive sensor array based on the single-frame capacitive image corresponding to the projected region; and using the area and coordinates as geometric features.

[0017] In the above technical solution, the spatial contour features include geometric features, which are the shape, size, and positional distribution characteristics of the projected area of ​​the target object on the projected capacitive sensor array. These geometric features can intuitively depict the spatial occupancy and orientation attributes of the target object from a spatial dimension, providing basic spatial feature support for subsequent target type discrimination and attitude recognition. Based on the single-frame capacitive image corresponding to the projected area, the area of ​​the projected area and the coordinates of the center of gravity of the projected area on the projected capacitive sensor array are calculated. The area and coordinates are used together as geometric features. The area can quantify the size of the projected coverage of the target object, effectively distinguishing target objects of different sizes. The center of gravity coordinates can locate the actual position of the target object in the sensing space, intuitively reflecting its spatial orientation distribution. The combination of the two can construct a spatial geometric representation of the target object, improving the reliability and accuracy of subsequent multi-dimensional feature analysis and target perception.

[0018] In conjunction with the first aspect, in some possible implementations, the spatial contour features also include: statistical features, which are statistical parameters calculated based on the capacitance values ​​within the projection region of the target object; the spatial contour features are calculated based on the spatial distribution of capacitance values ​​in a single-frame capacitance image of the projection region, including: calculating the mean and variance of capacitance values ​​within the projection region based on the single-frame capacitance image corresponding to the projection region; and using the mean and variance as statistical features.

[0019] In the above technical solution, the spatial contour features also include statistical features, which are statistical parameters calculated based on the capacitance values ​​within the projection area of ​​the target object. These statistical features can characterize the coupling characteristics between the target object and the projected capacitive sensor array from the perspective of capacitance value distribution, effectively supplementing the discriminative dimension of geometric features, which can only characterize spatial morphology. Based on the single-frame capacitance image corresponding to the projection area, the mean and variance of the capacitance values ​​within the projection area are calculated, and these mean and variance are used as statistical features. The mean can quantitatively characterize the overall average level of capacitive coupling strength within the projection area, intuitively reflecting the overall coupling strength between the target object and the sensor array. The variance can characterize the dispersion and uniformity of the capacitance values ​​relative to the mean, reflecting the spatial fluctuation differences in capacitance response. The combination of these two features can provide a stable statistical basis for target object type differentiation and posture recognition.

[0020] In conjunction with the first aspect, in some possible implementations, the spatial contour features also include: texture features, which are the image texture patterns of capacitance values ​​within the projection region of the target object; the spatial contour features are calculated based on the spatial distribution of capacitance values ​​in a single-frame capacitance image of the projection region, including: calculating the corresponding gray-level co-occurrence matrix based on the single-frame capacitance image corresponding to the projection region; the gray-level co-occurrence matrix is ​​a joint probability distribution matrix of pixel gray-level values ​​that appear in pairs in a preset direction and interval, based on the gray-level of the capacitance image; extracting contrast and energy parameters based on the gray-level co-occurrence matrix, where contrast is the degree of local gray-level change in the capacitance image; and energy parameters are the degree of gray-level distribution in the capacitance image; and using the contrast and energy parameters as texture features.

[0021] In the above technical solution, the spatial contour features also include texture features. These texture features are used to characterize the image texture pattern presented by the capacitance value within the projection area of ​​the target object. They can characterize the dielectric properties and surface morphology differences of the target object from the dimension of the texture distribution of the capacitance response, effectively making up for the limitations of geometric features that only describe spatial morphology and statistical features that only characterize numerical distribution, and supplementing the target perception with a new texture dimension. Based on the single-frame capacitance image corresponding to the projection area, the corresponding gray-level co-occurrence matrix is ​​calculated. This gray-level co-occurrence matrix is ​​a matrix formed by statistically analyzing the joint probability distribution of pixel gray values ​​that appear in pairs under preset directions and preset intervals, using the gray levels of the capacitance image as the statistical basis. It serves as the carrier for quantifying the texture pattern of the capacitance image. Then, contrast and energy parameters are extracted based on the gray-level co-occurrence matrix. The contrast parameter is used to characterize the degree of local gray-level change in the capacitance image, intuitively reflecting the fluctuation difference of the capacitance coupling strength and the roughness of the texture within the projection area. The energy parameter is used to characterize the regularity of the gray-level distribution of the capacitance image, corresponding to the uniformity of the target dielectric properties and the concentration of the texture distribution. Using the contrast parameter and energy parameter together as texture features can capture the specificity of different target objects in the capacitance response texture pattern, further enriching the representation dimension of spatial contour features.

[0022] In conjunction with the first aspect, in some possible implementations, a control strategy for the target device is generated based on multi-dimensional features, including: determining the attribute category of the target object based on multi-dimensional features; if the attribute category of the target object belongs to the protection category, then determining the spatial attitude of the target object based on multi-dimensional features; and generating a control strategy for the target device based on the attribute category or spatial attitude of the target object.

[0023] In the above technical solution, based on multi-dimensional features, the attribute category of the target object is determined, which can distinguish different types of target objects such as human occupants, child safety seats, and non-biological objects, thus achieving reliable classification and identification of the target object. If the attribute category of the target object is determined to belong to the protection category that requires safety protection, the spatial posture of the target object is further determined by multi-dimensional feature analysis. This can capture key spatial posture information such as occupant leaning forward, sitting sideways, and positional deviation, effectively making up for the limitations of single-dimensional feature recognition. Finally, by combining the attribute category and spatial posture of the target object, a target device control strategy that matches the current detection scenario is adaptively generated. This strategy can not only adjust the execution logic of the safety device according to the differences in the target category to avoid false triggering or missed triggering, but also match the optimal safety protection scheme for different spatial postures, greatly improving the adaptability and overall reliability of the target device safety control.

[0024] In conjunction with the first aspect, in some possible implementations, multi-dimensional features include: living features, which are signal features of the periodic physiological activities of living organisms; and determining the attribute category of the target object based on the multi-dimensional features, including: determining whether the target object has life characteristics based on the living features; if it has life characteristics, then determining the attribute category of the target object as a living organism; if it does not have life characteristics, then determining the attribute category of the target object as a non-living organism.

[0025] In the above technical solution, the multi-dimensional features include living features, which are signal features characterizing the periodic physiological activities of living organisms. These features can capture the periodic changes in capacitance caused by micro-movements such as breathing and heartbeat in the human body from the time domain dimension, effectively making up for the technical limitations of geometric features, statistical features, and texture features, which can only reflect spatial information and cannot distinguish life attributes. Based on these living features, it can be determined whether the target object has life characteristics. If the target object is determined to have life characteristics, its attribute category is determined to be a living organism. If the target object is determined not to have life characteristics, its attribute category is determined to be a non-living organism. This enables rapid initial classification of the life attributes of the target object and effectively eliminates interference from non-biological life-like organisms such as humanoid dolls and robots.

[0026] In conjunction with the first aspect, in some possible implementations, the multi-dimensional features also include: geometric features, which are the shape, size, and positional distribution characteristics of the projected area of ​​the target object on the projected capacitive sensor array; determining the attribute category of the target object based on the multi-dimensional features also includes: if the attribute category of the target object is a living organism, then based on the area and coordinates in the geometric features, determining whether the area is greater than or equal to a preset area threshold; if the area is greater than or equal to the preset area threshold, then determining the attribute category of the living organism is an adult; if the area is less than the preset area threshold and the coordinates meet the preset coordinate conditions, then determining the attribute category of the living organism is a child.

[0027] In the above technical solution, the multi-dimensional features also include geometric features. These geometric features characterize the shape, size, and positional distribution of the projected area of ​​the target object on the projected capacitive sensor array. They can depict the spatial occupancy and orientation information of the target object from a spatial dimension, providing a reliable basis for the refined subtype classification of organisms. Given that the target object's attribute category is determined to be an organism, based on the area and coordinate information in the geometric features, it is determined whether the area is greater than or equal to a preset area threshold. If the area is greater than or equal to the preset area threshold, the attribute category of the organism is directly determined to be an adult. This allows for efficient differentiation between adults and other organisms of different body types based on the difference in the projected body size, ensuring real-time classification. If the area is less than the preset area threshold and the coordinates meet the preset coordinate conditions, the attribute category of the organism is determined to be a child. This dual-judgment logic, combining initial screening with preset area thresholds and verification with preset coordinate conditions, effectively avoids misjudgments caused by crouching posture, positional shifts, or local interference, improving the accuracy of identifying adult and child organism subtypes.

[0028] In conjunction with the first aspect, in some possible implementations, the multi-dimensional features also include: statistical features, which are statistical parameters calculated based on the capacitance values ​​within the projected region of the target object; determining the attribute category of the target object based on the multi-dimensional features further includes: if the attribute category of the target object is a living organism, then based on the mean and variance in the statistical features, determining whether the mean is greater than a preset mean threshold and the variance is less than a preset variance threshold; if the mean is greater than the preset mean threshold and the variance is less than the preset variance threshold, then determining the attribute category of the living organism is an adult; if the mean is less than or equal to the preset mean threshold and the variance is greater than or equal to the preset variance threshold, then determining the attribute category of the living organism is a child.

[0029] In the above technical solution, the multi-dimensional features also include statistical features. These statistical features are statistical parameters calculated based on the capacitance values ​​within the projected area of ​​the target object. They can characterize the differences in coupling characteristics of the target object from the perspectives of the central tendency and discrete distribution of the capacitance response, effectively compensating for the limitations of single geometric features in biological subdivision scenarios and improving the anti-interference ability of adult and child classification. Based on the target object's attribute category already determined to be a biological object, a joint conditional judgment is performed based on the mean and variance in the statistical features. That is, it is determined whether the mean is greater than a preset mean threshold and the variance is less than a preset variance threshold. If both are true... If the above conditions are met, the attribute category of the organism is determined to be adult, based on the characteristics that adults have higher capacitive coupling strength with the projected capacitive sensor array and more uniform and stable capacitance value distribution. If the mean is less than or equal to a preset mean threshold and the variance is greater than or equal to a preset variance threshold, the attribute category of the organism is determined to be child, based on the characteristics that children have relatively lower capacitive coupling strength with the sensor array and higher dispersion of capacitance value distribution. By jointly judging the dual statistical parameters of mean and variance, the classification interference caused by differences in limb posture and sitting posture can be reduced, and stable distinction between adult and child organisms can be achieved.

[0030] In conjunction with the first aspect, in some possible implementations, the multi-dimensional features also include: texture features, which are the image texture patterns of capacitance values ​​within the projection area of ​​the target object; determining the attribute category of the target object based on the multi-dimensional features also includes: if the attribute category of the target object is a living organism, then based on the contrast and energy parameters in the texture features, determining whether the contrast is less than a preset contrast threshold and whether the energy parameter is greater than a preset energy threshold; the contrast is the degree of local grayscale change in the capacitance image; the energy parameter is the degree of grayscale distribution in the capacitance image; if the contrast is less than the preset contrast threshold and the energy parameter is greater than the preset energy threshold, then the attribute category of the living organism is determined to be an adult; if the contrast is greater than or equal to the preset contrast threshold and the energy parameter is less than or equal to the preset energy threshold, then the attribute category of the living organism is determined to be a child.

[0031] In the above technical solution, the multi-dimensional features also include texture features. These texture features are used to characterize the image texture pattern corresponding to the capacitance value within the projection area of ​​the target object. They can supplement the classification criteria of organisms from the dimension of the texture distribution of capacitance response, make up for the limitations of geometric and statistical features, and further improve the accuracy and anti-interference ability of adult and child classification. Contrast is used to reflect the degree of local grayscale change in the capacitance image, and the energy parameter is used to characterize the regularity of the grayscale distribution in the capacitance image. Based on the fact that the attribute category of the target object has been determined to be an organism, a joint conditional judgment is performed based on the contrast and energy parameter in the texture features, i.e., determining whether the contrast is less than a preset value. If the contrast threshold and energy parameter are both greater than the preset energy threshold, and the above conditions are met, then the attribute category of the organism is determined to be adult, based on the characteristics that the grayscale change of the capacitance response generated by the interaction between the adult and the sensor array is smooth and the grayscale distribution is more concentrated and uniform. If the contrast is greater than or equal to the preset contrast threshold and the energy parameter is less than or equal to the preset energy threshold, then the attribute category of the organism is determined to be child, based on the characteristics that the grayscale fluctuation of the capacitance response is more significant and the grayscale distribution is more discrete. Through the collaborative discrimination of contrast and energy dual texture parameters, the classification error caused by sitting posture changes, limb occlusion and other scenarios can be effectively reduced, and the refined identification of organism subtypes can be achieved.

[0032] In conjunction with the first aspect, in some possible implementations, before determining the spatial pose of the target object based on multi-dimensional features, the method further includes: determining the attribute category of the target object based on a preset category protection mapping table; if the attribute category belongs to a category in the preset category protection mapping table, then the attribute category is determined to belong to a protection category; if the attribute category does not belong to a category in the preset category protection mapping table, then the attribute category is determined not to belong to a protection category.

[0033] In the above technical solution, the attribute category of the target object is matched and determined based on a preset category protection mapping table. If the attribute category is determined to belong to a protected object category pre-recorded in the preset category protection mapping table, then the attribute category is clearly identified as a protection category, which can quickly identify the target type that requires targeted security protection for the target device. If the attribute category is determined not to be included in the preset category protection mapping table, then the attribute category is determined not to belong to a protection category, and there is no need to trigger the corresponding security protection mechanism. Through the standardized mapping table matching method, the logic of category classification can be simplified, the identification efficiency can be improved, and the identification error caused by manual or fuzzy judgment can be effectively avoided.

[0034] In conjunction with the first aspect, in some possible implementations, if the attribute category of the target object belongs to the protection category, the spatial pose of the target object is determined based on multi-dimensional features, including: if the attribute category of the target object belongs to the protection category, then the shape parameters of the projection area of ​​the target object on the projected capacitive sensor array are calculated based on the geometric features in the multi-dimensional features; the geometric features are the shape, size, and positional distribution characteristics of the projection area of ​​the target object on the projected capacitive sensor array; the spatial pose type of the target object is determined based on the coordinates and shape parameters of the geometric features; the coordinates of the geometric features are the coordinates of the centroid of the projection area on the projected capacitive sensor array.

[0035] In the above technical solution, if the attribute category of the target object belongs to the protection category requiring security protection, the shape parameters of the projection area of ​​the target object on the projected capacitive sensor array can be calculated based on the geometric features in the multi-dimensional features. The geometric features are used to characterize the shape, size, and position distribution of the projection area of ​​the target object on the projected capacitive sensor array, which can characterize the projection contour and spatial occupancy characteristics of the target object from the spatial dimension, providing a stable morphological data foundation for attitude recognition. Furthermore, based on the coordinates of the center of gravity of the projection area on the projected capacitive sensor array in the geometric features and the obtained shape parameters, the spatial attitude type of the target object is comprehensively determined. The center of gravity coordinates reflect the spatial position and offset state of the target object, and the shape parameters quantify the contour deformation characteristics of the projection area. The collaborative judgment of the two can improve the accuracy and anti-interference of spatial attitude determination and avoid the recognition deviation caused by a single feature.

[0036] In conjunction with the first aspect, in some possible implementations, the spatial posture type of the target object is determined based on the coordinates and shape parameters of the geometric features, including: if the coordinate value of the first axis in the coordinates of the geometric features is greater than the first posture threshold, then the spatial posture type of the target object is determined to be a forward-leaning posture; if the difference between the shape parameter and the reference orientation angle is greater than the second posture threshold, then the spatial posture type of the target object is determined to be a side-sitting posture; if neither of the conditions for determining the forward-leaning posture or the side-sitting posture is met, then the spatial posture type of the target object is determined to be a standard posture.

[0037] In the above technical solution, when determining the spatial posture type of the target object, if the first axis coordinate value of the centroid coordinate in the geometric feature is greater than the preset first posture threshold, the spatial posture type of the target object is determined to be a forward-leaning posture based on the coordinate offset feature. This can quickly identify the dangerous posture of the occupant's body leaning forward, providing timely judgment basis for the timing and intensity control of subsequent safety devices. If the difference between the shape parameter of the projection area and the preset reference direction angle is greater than the preset second posture threshold, the spatial posture type of the target object is determined to be a side-sitting posture based on the contour angle offset feature. This can capture the abnormal state of the occupant's side-sitting posture, make up for the blind spot of single coordinate judgment, and improve the completeness of posture recognition. If neither the determination conditions for forward-leaning posture nor side-sitting posture are met, the spatial posture type of the target object is determined to be a standard posture. Through the dual-layer joint judgment logic of coordinate threshold and shape angle difference, various typical occupant postures can be distinguished, reducing the probability of posture misjudgment and improving the reliability and robustness of spatial posture recognition.

[0038] In conjunction with the first aspect, in some possible implementations, a control strategy for the target device is generated based on the attribute category or spatial posture of the target object, including: if the attribute category of the target object is a non-living object, then an inhibition command is generated for the target device, which is used to inhibit the protective action of the safety device in the target device; if the attribute category of the target object is a child, then a first control command is generated for the target device, which is used to trigger the target device to control with a first intensity parameter, which is lower than a calibrated intensity parameter; if the attribute category of the target object is an adult and the corresponding spatial posture is a standard posture, then a second control command is generated for the target device, which is used to trigger the target device to control with a calibrated intensity parameter; if the attribute category of the target object is an adult and the corresponding spatial posture is a forward-leaning posture, then a third control command is generated for the target device, which is used to trigger the target device to control with a calibrated intensity parameter and control the adjustment direction.

[0039] In the above technical solution, when the target object is determined to be a non-living object, an inhibition command for the target device is automatically generated. This inhibition command restricts and blocks the protective triggering and execution of the safety device within the target device, avoiding the problem of accidental activation or triggering of the safety device in unoccupied scenarios, and mitigating unnecessary safety hazards and equipment damage. When the target object is determined to be a child, a first control command corresponding to the target device is generated. This first control command drives the target device to complete safety control according to a first intensity parameter. This first intensity parameter is pre-set to be lower than the standard calibration intensity parameter, which can adapt to the physical tolerance characteristics of children, reduce the operating intensity of the safety device, and prevent high-intensity protective actions from causing harm. Children may cause secondary injuries. When the target is an adult and the target is in a standard spatial posture, a second control command is generated to drive the safety device to work stably with standard calibration strength parameters, ensuring that the protective performance meets the standards under normal sitting conditions and providing balanced and reliable standard safety protection for adult occupants. When the target is an adult and is in a dangerous forward-leaning posture, a third control command is generated. While maintaining the calibration strength parameters unchanged and ensuring basic protection capabilities, the control direction of the safety device is adjusted simultaneously to specifically adapt to the special forward-leaning driving posture, making up for the limitations of protection in fixed protection directions and improving the protection accuracy, adaptability and operational safety of the passive safety system of the target equipment in complex driving scenarios.

[0040] Secondly, a control device is provided, comprising: an acquisition device for acquiring a corresponding capacitance image sequence in response to a target object entering the sensing space of a projected capacitance sensor array of a target device, the capacitance image sequence including multiple images showing the change in capacitance values ​​at various coordinate points on the projected capacitance sensor array over time caused by the target object; an analysis device for analyzing multi-dimensional features from the capacitance image sequence, the multi-dimensional features being features analyzed from the temporal and spatial domains of the capacitance image sequence; and a generation device for generating a control strategy for the target device based on the multi-dimensional features.

[0041] Thirdly, a vehicle is provided, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, causing the vehicle to perform the methods described in the first aspect or any possible implementation thereof.

[0042] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application so as to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more easily understood, specific embodiments of this application are given below. Attached Figure Description

[0043] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic diagram of the structure of a vehicle provided in an embodiment of this application; Figure 2 A flowchart illustrating a control method provided in an embodiment of this application. Figure 1 ; Figure 3 A flowchart illustrating a control method provided in an embodiment of this application. Figure 2 ; Figure 4 A flowchart illustrating a control method provided in an embodiment of this application. Figure 3 ; Figure 5 This is a schematic diagram of the structure of a control device provided in an embodiment of this application. Detailed Implementation

[0044] The technical solutions of this application will now be described clearly and in detail with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " signifies "or," for example, A / B can mean A or B. "And / or" in the text merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more.

[0045] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0046] In the safety control of target equipment (such as vehicles), the perception of internal target objects is crucial. In the vehicle field, existing occupant classification systems mostly use seat-embedded pressure sensor matrices or strain gauges to distinguish target object types by collecting weight data or pressure distribution, thereby providing information to the control unit to generate safety control strategies. However, this technology can only acquire single-dimensional perception information and cannot accurately distinguish between target objects of similar weight but different types (such as children and heavy objects) or target objects of the same size but different types (such as real people and humanoid dolls or humanoid robots) in complex scenarios. Furthermore, it cannot perceive the vital signs and key spatial postures of target objects (such as leaning forward or sitting sideways). This lack of perception dimensions not only easily leads to safety hazards due to misclassification but also makes it difficult to effectively implement safety control strategies due to the lack of necessary input information. Therefore, there is an urgent need for a technical solution that can solve the above-mentioned technical problems, and a control scheme is proposed accordingly.

[0047] The control method and vehicle of the present application will be described in detail below with reference to the accompanying drawings and through multiple embodiments.

[0048] Figure 1 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application. Figure 1 As shown, the vehicle 100 may include a processor 110 and a memory 120.

[0049] The memory 120 stores machine-executable instructions that can be executed by the processor 110. When the vehicle 100 is running, these machine-executable instructions are executed. The processor 110 and the memory 120 communicate via a bus. The processor 110 can execute these machine-executable instructions to implement the control method.

[0050] The memory 120, processor 110, and various bus components are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 120 includes at least one software functional module, which is stored or embedded in the vehicle's operating system (OS) in the form of software or firmware. This software functional module includes at least one executable module. The processor 110 is used to execute the executable modules stored in the memory 120, such as the software functional modules and computer programs included in the control method.

[0051] The memory 120 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0052] The vehicle 100 can be selected according to the actual situation; for example, the vehicle 100 may have software that can execute control methods.

[0053] The control method provided in this application embodiment can be executed by a processor in the vehicle 100. The control method provided in this application embodiment will be explained further below. Figure 2 A flowchart illustrating a control method provided in an embodiment of this application. Figure 1 .like Figure 2 As shown, the method may include: S210, In response to the target object entering the sensing space of the projected capacitive sensor array of the target device, acquire the corresponding capacitance image sequence.

[0054] The capacitance image sequence comprises multiple images showing the change in capacitance values ​​at various coordinate points on the projected capacitance sensor array over time, caused by the target object. In other words, the capacitance image sequence is a time-series collection of capacitance data recording the entire sensing process of the target object.

[0055] Among them, each coordinate point on the projected capacitive sensor array refers to the pre-calibrated matrix electrode coordinates. Each matrix electrode coordinate uniquely corresponds to a physical sensing electrode on the projected capacitive sensor array, providing a precise spatial position mapping relationship for the capacitance value.

[0056] When a target object enters the sensing space, it forms a coupling capacitance with the electrodes of the projected capacitive sensor array based on the principle of capacitive coupling. This alters the inherent capacitance value of each electrode. Because the relative distance and contact area between the target object and different coordinate electrodes vary, the magnitude and trend of capacitance changes at each coordinate point differ. These capacitance changes are directly triggered by the target object's sensing behavior. "Change over time" refers to the sensing unit continuously sampling at fixed time intervals. Each frame of the capacitance image corresponds to an independent timestamp. As the target object's position and posture dynamically change within the sensing space, the capacitance values ​​at each coordinate point exhibit continuous temporal changes. Multiple images refer to multiple frames of capacitance images acquired sequentially over time. Each frame is a two-dimensional digital mapping of the capacitance values ​​at all coordinate points of the projected capacitive sensor array at the corresponding moment. These multiple frames are combined chronologically to form a sequence, preserving the spatiotemporal characteristics of the capacitance changes caused by the target object. This provides a traceable sensing data foundation for subsequent processing of the target object's trajectory and posture.

[0057] It should be noted that the projected capacitive sensor array is essentially a projected capacitive two-dimensional matrix sensor array, belonging to high-precision capacitive sensing components. It consists of several sets of transverse transmitting electrodes (Tx) and longitudinal receiving electrodes (Rx) arranged in an orthogonal matrix to form an array grid. Each intersection point of the transmitting and receiving electrodes corresponds to an independent spatial coordinate. During operation, the transmitting electrodes continuously output a high-frequency excitation electric field, and the receiving electrodes sense the coupled electric field transmitted from the transmitting electrodes, thereby forming a stable reference mutual capacitance at the electrode intersection point. The array pre-defines a three-dimensional sensing space covering the area above the electrodes. When a target object (conductive or dielectric object) enters this three-dimensional sensing space, it will disturb the distribution of the coupled electric field between the electrodes, directly causing the mutual capacitance values ​​at each coordinate intersection point to change regularly. The array can simultaneously collect the capacitance changes at all coordinate points and convert them into two-dimensional capacitance distribution data. It is a dedicated array sensor capable of realizing full-area point capacitance detection and outputting capacitance images, and it is also the hardware foundation for acquiring capacitance image sequences. Furthermore, it should be noted that ordinary sensor arrays mostly employ single-electrode capacitance detection, resistance deformation detection, or piezoelectric signal detection, enabling only single-point physical quantity acquisition. In contrast, projected capacitance sensor arrays rely on the mutual capacitance coupling electric field between electrodes to achieve non-contact detection through electric field disturbances. Secondly, their spatial detection capabilities differ. Ordinary sensor arrays lack three-dimensional sensing space, can only detect contact signals with low point detection accuracy, and cannot form a global spatial perception. Projected capacitance sensor arrays can cover a three-dimensional sensing space, capturing capacitance changes caused by the position and posture of the target object within the space. Thirdly, their output data formats differ. Ordinary sensor arrays can only output discrete single-point analog or digital quantities, unable to form image data. Projected capacitance sensor arrays can integrate the capacitance values ​​of each coordinate point into a two-dimensional capacitance image, and continuous acquisition can form a time-series capacitance image sequence. Finally, their detection accuracy and functionality differ. Ordinary sensor arrays are susceptible to environmental interference and only support coarse single-point trigger detection. Projected capacitance sensor arrays have strong anti-interference capabilities, enabling high-precision global point detection, multi-point synchronous detection, and non-contact spatial perception, thus meeting the technical requirements for acquiring capacitance image sequences.

[0058] The target device is a terminal device integrating projected capacitive sensing detection function. Its type and form can be flexibly selected and configured according to the actual application scenario, including but not limited to vehicles (passenger cars, commercial vehicles, and other in-vehicle interactive devices), smart home control terminals, smart cockpit devices, portable electronic devices, and smart office devices. The target device has a built-in MCU (Microcontroller Unit) and a projected capacitive sensor array. The projected capacitive sensor array is deployed in the interactive operation area of ​​the target device to sense and collect capacitive signals from the approach, movement, and contact behavior of external target objects, so as to realize non-contact or contact intelligent detection and interactive control.

[0059] The target object is a physical object that can generate capacitive coupling with the projected capacitive sensor array and cause a detectable change in the capacitance value at each coordinate point of the sensor array. This includes, but is not limited to, human body parts (such as user fingers, palms, and arms) and physical objects with conductive or dielectric properties. The target object can enter the sensing space of the projected capacitive sensor array. By changing the electric field distribution and coupling capacitance between the electrodes, the projected capacitive sensor array outputs a recognizable capacitance change signal, which is then used as the detection object generated by the detection, feature extraction, and control strategies.

[0060] In one possible implementation, because the projected capacitive sensor array consists of several transmitting and receiving electrodes arranged in a matrix, with each electrode point corresponding to a unique spatial coordinate, this projected capacitive sensor array pre-exists a sensing space. This space is a three-dimensional effective detection area where the electrodes effectively capacitively couple with external conductive or dielectric targets, and changes in capacitance can be detected. The target object is an object that can trigger capacitive coupling. Therefore, the action of the target object entering this sensing space will disrupt the preset reference capacitance balance state of the projected capacitive sensor array. That is, the projected capacitive sensor array continuously monitors the reference capacitance value (i.e., the default capacitance value) of each electrode coordinate point at a constant scanning frequency. When a jump exceeding a preset capacitance threshold is detected in the capacitance value of any coordinate point compared to the reference capacitance value, it is determined that the target object has entered the sensing space, triggering a data acquisition response. At this time, the sensing and acquisition unit reads the real-time capacitance values ​​of all coordinate points on the projected capacitive sensor array row by row and column by column according to the preset time sequence sampling rules. The capacitance values ​​of all coordinate points at the same sampling time are integrated into a single frame capacitance image according to the array coordinate arrangement relationship. Multiple frames of this type of capacitance image are continuously acquired in chronological order, and finally a capacitance image sequence corresponding to the sensing action of the target object is formed. The entire process ensures time synchronization and one-to-one correspondence with coordinates, and completes the acquisition of the capacitance image sequence.

[0061] S220. Extract multi-dimensional features from the capacitance image sequence.

[0062] The multi-dimensional features are a set of features extracted from the temporal and spatial domains of the capacitance image sequence, reflecting the law of capacitance change caused by the target object. They also represent the target object's operational behavior, state, and the operating conditions of the target device, specifically encompassing spatial morphology, dynamic processes, physical attributes, and operational intentions. These can be divided into three main categories: First, capacitance change features, extracted by performing differential operations and threshold filtering on the capacitance values ​​of each frame of the capacitance image sequence. These reflect the coupling strength and dynamic changes between the target object and the sensor array, and include at least: the capacitance change amplitude, capacitance change rate, and capacitance steady-state value deviation at each coordinate point of the projected capacitance sensor array. Second, target object spatial features, calculated by combining the distribution law of capacitance change and coordinate mapping relationship in the capacitance image sequence with a preset spatial calibration algorithm. These are used to locate the position and spatial morphology of the target object, and include at least: the spatial coordinates, contact range, and attitude angle of the target object within the sensor array's sensing space. Thirdly, there are operational intent features. These combine the aforementioned capacitance change features with the spatial features of the target object, using time-series analysis and pattern matching to identify the target object's operational actions (such as approaching, moving away, staying, sliding, etc.), and then parsing the corresponding operational intent (such as starting the device, adjusting parameters, turning off the function, etc.). These three types of features together constitute multi-dimensional features, providing a reliable decision-making basis for subsequent steps.

[0063] From the perspective of feature analysis dimensions, these multi-dimensional features can be divided into spatial domain features, temporal domain features, and amplitude features. These three types of features do not overlap and complement each other. Among them, spatial domain features are spatially distributed features obtained from single-frame capacitance images of the capacitance image sequence. They focus on the spatial position and contact range of the target object within the sensing space of the projected capacitance sensor array at a single moment. These features include at least: shape features of the capacitance change region (such as circular, rectangular, or irregular shapes), area features (i.e., the number of coordinate points where the capacitance value changes significantly), centroid coordinate features (i.e., the sensor array coordinates corresponding to the geometric center of the capacitance change region), and edge contour features (i.e., the set of boundary coordinates between the capacitance change region and the non-change region). Temporal features are time-dependent features derived from multiple consecutive frames of a capacitance image sequence. They focus on the dynamic changes of the target object within the sensing space and include at least: capacitance change rate (the ratio of the difference in capacitance at the same coordinate point between two adjacent frames to the time interval), capacitance peak characteristics (the maximum capacitance change at a certain coordinate point across multiple frames), capacitance change periodic characteristics (the periodicity of capacitance changes caused by repeated actions of the target object), and capacitance steady-state characteristics (the stable range of capacitance when the target object is stationary). Based on the amplitude and distribution gradient of the capacitance change at each coordinate point relative to the reference value, amplitude-dimensional features such as dielectric properties, coupling strength, and distance of the target object can also be analyzed. These spatial, temporal, and amplitude-dimensional features reflect the sensing characteristics of the target object from the perspectives of spatial morphology, dynamic process, and physical properties, respectively.

[0064] It should be noted that the parsing processes of spatial domain features and temporal domain features are independent yet coordinated. Spatial domain features reflect the static spatial attributes of the target object, while temporal domain features reflect the dynamic temporal attributes of the target object. The multi-dimensional features formed by the combination of the two can capture the core features of the target object and provide reliable feature support for subsequent steps.

[0065] In one possible implementation, since the capacitance image sequence is a collection of multiple frames of capacitance images acquired sequentially in chronological order, each frame contains the real-time capacitance values ​​of all coordinate points of the projected capacitance sensor array, and each frame corresponds to a unique timestamp, recording the spatiotemporal process of capacitance change triggered by the target object entering the sensing space. Therefore, during the analysis process, the capacitance image sequence is first preprocessed to eliminate the influence of environmental interference (such as temperature drift and electromagnetic noise) and acquisition errors. This preprocessing operation can be performed to reduce noise in each frame of capacitance image through mean filtering and Gaussian denoising algorithms, and then the capacitance values ​​of each coordinate point are mapped to a unified quantization range (such as [0,255]) through a normalization algorithm to ensure the comparability of capacitance data of different frames and different coordinate points. After the preprocessing is completed, a preset feature extraction algorithm is called to extract features in both the temporal and spatial domains from the capacitance image sequence. Among them, spatial features are extracted from the spatial distribution of a single-frame capacitance image, while temporal features are extracted from the temporal variation trend of multiple-frame capacitance images. By fusing multi-dimensional features, the capacitance variation pattern of the target object is captured, avoiding information loss caused by single-dimensional features, and ensuring that the parsed features are unique, stable, and discriminative, thus meeting the requirements of subsequent technical processes for feature data.

[0066] It should be noted that the capacitance image sequence is not a single-dimensional set of values, but a composite data that simultaneously carries three types of basic physical information: spatial distribution, temporal variation, and capacitance amplitude variation. Each frame of the capacitance image corresponds to the spatial distribution characteristics of capacitance at each coordinate point of the sensor array. Multiple consecutive frames of images record the dynamic process of capacitance change over time. By combining the magnitude and gradient differences of capacitance change at each coordinate point, non-overlapping and complementary multi-dimensional features can be extracted. Based on this composite data, multi-dimensional features covering multiple levels such as space, time, amplitude, and operational intent can be analyzed, providing feature support for subsequent processing.

[0067] S230. Based on multi-dimensional features, generate control strategies for the target device.

[0068] In one possible implementation, multi-dimensional features can objectively characterize the interactive behavior, true operational intent, and current application scenario status information of the target object from multiple levels, such as spatial distribution, temporal changes, behavioral patterns, and signal characteristics. This compensates for the shortcomings of single feature information, such as being one-sided, susceptible to interference, and unable to reflect actual operational needs. Furthermore, the control system of this application pre-constructs a quantitative mapping relationship and decision rules between multi-dimensional feature combinations and target device control logic. It can match and judge the intent information represented by multi-dimensional features with a preset control rule library. At the same time, multi-dimensional features can mutually verify each other and eliminate abnormal interference, greatly improving the reliability of intent recognition and status determination. This provides the control system with sufficient and reliable decision-making basis, enabling the control system to select execution logic that is highly adapted to the current state and operational intent. Ultimately, it generates a scientific and executable target device control strategy, realizing intelligent and adaptive control of the target device. Based on this, the multi-dimensional features are first standardized according to a preset data format to remove redundant data, correct abnormal feature values, and ensure the validity and consistency of the feature data. Then, the standardized multi-dimensional features are matched one by one with various rules in the preset control rule base to select the control logic that best matches the current multi-dimensional features. If multiple sets of matching control rules exist, the optimal control logic is selected through a priority ranking algorithm (such as weighted ranking based on feature matching degree). Finally, a corresponding control strategy is generated based on the optimal control logic. This control strategy is a standardized set of instructions that specifies the execution actions, execution parameters, and execution timing of the target device (e.g., "start the core function of the target device, adjust the operating power to 50%, and execute after a 2-second delay"). After generation, the target device's execution unit (such as the drive module, display module, etc.) executes the strategy, and the corresponding multi-dimensional features are simultaneously stored in the target device's storage unit for subsequent strategy optimization and fault tracing.

[0069] The preset control rule library stores the correspondence between different multi-dimensional feature combinations and target device control commands, covering the control logic under various operating modes and scenarios of the target device. For example, when the capacitance change of the target object exceeds the preset threshold and the spatial coordinates are located in the core area of ​​the sensor array, the corresponding control logic is "start the target device"; when the target object's attitude angle is continuously changed and the capacitance change rate is stable, the corresponding control logic is "adjust the target device parameters".

[0070] For example, taking a scenario where only a single central control display screen is configured in the front row of a vehicle, even if the central control display screen is not directly facing the driver or passenger, the control system can still detect the target object (such as a passenger or external object) when it enters the sensing space of the projected capacitive sensor array mounted on the central control display screen. The principle behind this is that the sensing space of the projected capacitive sensor array is a three-dimensional effective capacitive coupling area covering the periphery of the sensor, not limited to the narrow space directly opposite the normal direction of the projected capacitive sensor array. As long as the target object enters this three-dimensional sensing space, it can form a stable capacitive coupling with the transmitting and receiving electrodes in the projected capacitive sensor array, thereby causing a detectable and identifiable change in the reference capacitance value of each coordinate point on the projected capacitive sensor array, thus completing the detection and determination of the target object. In conventional deployment and usage scenarios, when the projected capacitive sensor array of the target object and the target device is in a facing orientation, the coupling distance between the target object and the projected capacitive sensor electrodes is more uniform, the effective coupling area is larger, the resulting capacitive coupling strength is higher, the corresponding capacitance value change is more significant, and the noise of the projected capacitive sensing signal is lower. Therefore, under this orientation, the projected capacitive sensor array has the best effect on the capacitance image acquisition of the target object, which can provide higher quality and more stable raw sensing data for subsequent multi-dimensional feature extraction and control strategy generation.

[0071] For example, taking a smart seat with an integrated projected capacitive sensor array as the target device, the projected capacitive sensor array is arranged in a matrix on the seat surface, with each electrode corresponding to a unique spatial coordinate. The preset control rule base also pre-stores feature templates for different types of target objects, capacitance fluctuation thresholds corresponding to vital signs, and capacitance distribution templates for different postures to ensure the feasibility of differentiation and perception.

[0072] For target objects of similar weight but different types (such as a 30kg child and a 30kg object filled with books), the dielectric properties of their sensing areas are similar, resulting in essentially the same static capacitance change amplitude at each coordinate point of the projected capacitive sensor array. Therefore, relying solely on the steady-state capacitance value corresponding to the weight is insufficient to distinguish between these target objects. Therefore, this application uses a projected capacitive sensor array to acquire capacitance image sequences of both objects at a high-frequency sampling frequency of 100Hz. After preprocessing the capacitance image sequences, the capacitance change characteristics and spatial-temporal characteristics of each object are extracted. Children, as living subjects, exhibit subtle bodily movements due to breathing and heartbeats. This causes periodic, minute fluctuations in the capacitance values ​​at corresponding coordinate points in the projected capacitive sensor array (e.g., fluctuation frequency of 0.2-0.5Hz, fluctuation amplitude of 5-10pF). Furthermore, children naturally adjust their posture while seated (e.g., slight swaying, body shifting). Consequently, the capacitance values ​​at each coordinate point in the corresponding capacitance image sequence show dynamic temporal changes over time. In contrast, a 30kg heavy object, lacking living characteristics, maintains a steady state in its capacitance image sequence, exhibiting no periodic fluctuations and a fixed capacitance distribution without dynamic temporal changes. By matching the extracted capacitance change characteristics and spatial temporal features of each object with the preset control rule base of "capacitance fluctuation thresholds corresponding to living characteristics" and "capacitance temporal feature templates for static heavy objects," the difference between children and heavy objects can be distinguished.

[0073] For target objects of similar size but different types (such as an adult and a humanoid doll with a height of about 170cm and similar build), the contact area and capacitance distribution spatial characteristics formed by their interaction with the sensor array are basically the same. Relying solely on the capacitance characteristics in the spatial dimension is insufficient to identify and distinguish between the two types of target objects. To address this problem, this application introduces and integrates vital sign characteristics and dynamic posture characteristics to achieve differentiated discrimination. By utilizing the detectable vital signs and natural and consistent dynamic postures of the human body, a clear distinguishing criterion is formed between the lifeless characteristics and the fixed and rigid postures of humanoid dolls, thereby achieving effective identification between adults and humanoid dolls. Specifically, an adult's breathing and heartbeat cause periodic capacitance fluctuations in the sensor array. Furthermore, in a seated position, natural muscle micro-movements and posture adjustments (such as head turning and slight body tilting) also occur. These characteristics manifest as periodic fluctuation signals and dynamic capacitance changes in the capacitance image sequence. Humanoid dolls, lacking vital signs, do not exhibit periodic capacitance fluctuations in their corresponding capacitance image sequences. Even when artificially positioned in a fixed posture, their capacitance distribution remains constant, without any temporal capacitance changes caused by natural posture adjustments. By detecting the presence of capacitance fluctuations corresponding to vital signs and combining this with the dynamic change characteristics of the capacitance temporal features, it is possible to quickly and accurately distinguish between real humans and humanoid dolls.

[0074] It should be noted that the distinction between humanoid robots and real people is similar. Since humanoid robots have no vital signs and their posture adjustments are mechanical movements, the corresponding capacitance changes are unnatural, intermittent jumps, which are significantly different from the natural dynamic capacitance changes of real people. The two can be further distinguished by using a preset capacitance feature template of mechanical movements.

[0075] For sensing the key spatial posture of a target object, the state can be determined through the multi-dimensional features of the capacitance image sequence. In terms of vital sign sensing, by filtering the capacitance image sequence to remove noise signals caused by environmental interference, the periodic fluctuation signal is extracted. If the frequency and amplitude of the fluctuation signal are within the preset vital sign threshold range, that is, the frequency corresponds to a respiratory rhythm of 0.2-0.5Hz or a heart rhythm of 1-1.5Hz, and the amplitude is in the range of 5-10pF, it can be determined that the target object has vital signs.

[0076] In terms of spatial posture perception, such as leaning forward and sitting sideways, identification is achieved by extracting spatial features and capacitance distribution features. When the target object is leaning forward, its upper body is close to the front of the seat, which increases the capacitance change amplitude and expands the contact area at the front coordinate point of the projected capacitive sensor array, while the capacitance change amplitude at the rear coordinate point decreases accordingly. By extracting the regional difference features of capacitance distribution and matching them with a preset forward-leaning posture capacitance distribution template, it can be determined that the target object is leaning forward. When the target object is sitting sideways, its center of gravity is biased to one side of the seat, and the capacitance change amplitude at the coordinate point of the projected capacitive sensor array on the corresponding side increases and the contact area is concentrated, while the capacitance change amplitude on the other side is smaller. By detecting the asymmetry of capacitance distribution and the concentrated area of ​​contact area, combined with a preset side-sitting posture capacitance distribution template, it can be determined that the target object is sitting sideways. At the same time, by combining the capacitance temporal change features, the dynamic process of posture adjustment can be further identified to ensure the accuracy of posture perception.

[0077] The control method provided in this application acquires a capacitance image sequence composed of multiple frames of images showing the capacitance values ​​of each coordinate point of the sensor array changing over time in response to a target object entering the sensing space of the projected capacitive sensor array of the target device. This abandons the inherent mode of existing solutions that rely solely on single-dimensional information such as weight data and pressure distribution for perception, thus expanding the dimensions of information acquisition from the source of perception. Furthermore, multi-dimensional features are parsed from the temporal and spatial domains of this capacitance image sequence. With the help of these spatiotemporal features, it can effectively distinguish target objects with similar weights but different types, or similar body shapes but different categories. It can also perceive the vital signs, leaning forward, or sitting sideways of the target object, solving the safety hazard of target misclassification caused by insufficient single-dimensional perception. Finally, a control strategy for the target device is generated based on these multi-dimensional features, providing sufficient and reliable input information for the generation of a safety control strategy. This overcomes the technical problem of existing technologies being unable to achieve effective safety control strategies due to insufficient sensing information, improving the reliability of target object identification and the safety of target device control in complex scenarios.

[0078] Figure 3 A flowchart illustrating a control method provided in an embodiment of this application. Figure 2 .like Figure 3 As shown, the above method extracts multi-dimensional features from the capacitance image sequence, including: S310. Process the capacitance image sequence to determine the projection area of ​​the target object on the projected capacitance sensor array.

[0079] In one possible approach, a preprocessing operation is first performed on the capacitance image sequence. A pre-defined mean or median filtering algorithm is used to remove abnormal capacitance values ​​caused by environmental interference, circuit noise, and other factors, ensuring the accuracy and reliability of the original sensor data. Subsequently, temporal difference calculation is performed on multiple consecutive frames of capacitance images within the preprocessed sequence. This involves calculating the difference between the real-time detected capacitance value at each coordinate point and the pre-calibrated reference static capacitance value of the projected capacitance sensor array to obtain the capacitance change at each coordinate point. Based on this, a weighted fusion process is performed on the capacitance changes across multiple frames to remove spurious capacitance changes caused by transient interference, retaining only the effective capacitance change distribution data generated by the coupling effect of the target object. This completes the standardized sequence processing flow of noise reduction, temporal difference, and weighted fusion for the capacitance image sequence, providing a stable and reliable data foundation for the subsequent positioning of the target object's projection area.

[0080] Based on the standardized processing of the capacitance image sequence, a calibration operation is performed on the projected region of the target object using the obtained effective capacitance change distribution data. Since each coordinate point on the projected capacitance sensor array corresponds to a unique physical sensing electrode, the target object, upon entering the sensing space, will form a capacitive coupling effect with the array electrodes, resulting in detectable capacitance changes only at electrode points within a local area. Therefore, coordinate points with capacitance changes exceeding a preset capacitance threshold are first selected as effective sensing points. Then, a connected component clustering algorithm is used to spatially aggregate these effective sensing points, grouping those with adjacent spatial locations and similar capacitance change characteristics into the same data set. The continuous coordinate range covered by this data set is the projected region of the target object on the projected capacitance sensor array. This projected region is a two-dimensional sensing region characterizing the spatial coverage of the target object on the sensor array detection plane. Its boundary coordinates are determined by the extreme coordinate points in the clustering results. This calibration method distinguishes the coupling region of the target object from the interference-free static background region, achieving a unique determination of the target object's projected region and providing a spatial range basis for subsequent multi-dimensional feature extraction and control strategy generation.

[0081] It should be noted that, because the capacitance change of a projected capacitive sensor array has a strict physical correspondence with the spatial position of the target object—that is, the projected capacitive sensor array is composed of inductive electrodes arranged in rows and columns, each electrode point corresponds to a unique coordinate and has a stable reference capacitance value—when the target object, as a conductive or dielectric object, enters the sensing space, it will form capacitive coupling with the array electrodes. Only the electrode points covered by the vertical projection of the target object will produce detectable capacitance changes, while the electrode capacitance values ​​in non-projected areas will remain essentially at the reference state without significant fluctuations. When processing the capacitance image sequence, the real-time capacitance value of a single frame capacitance image is first compared with the reference capacitance value. By performing differential calculations, the capacitance change at each coordinate point is extracted, and noise interference and invalid weak changes are eliminated. Then, the valid coordinate points exceeding the preset capacitance threshold are spatially clustered and bounded. At the same time, the continuous features of multiple frames of temporal capacitance images are combined to filter out instantaneous interference and lock the stable capacitance change region. In the coordinate system of the projected capacitance sensor array, this capacitance change region coincides with the vertical projection range of the target object on the surface of the projected capacitance sensor array. Therefore, through the above differential calculation, noise filtering, threshold determination and temporal fusion processing of the capacitance image sequence, the projection area of ​​the target object on the projected capacitance sensor array can be determined.

[0082] For example, using an adult as the target and a vehicle as the target device: the adult does not enter the sensing area of ​​the projected capacitive sensor array in a vertically facing posture, but rather enters the sensing space from the side in an tilted, sideways posture. The dielectric properties of the adult's body will form a diagonally distributed capacitively coupled projection area on the sensing plane of the projected capacitive sensor array. This projection area is not a regular circle or a symmetrical shape, but rather exhibits an asymmetrical distribution that extends with the tilt angle of the body. Due to the different relative distances between different coordinate points on the projected capacitive sensor array and the human body, each point will generate a gradient change in capacitance value. Ultimately, this results in an asymmetrical capacitance bright spot distribution in a single frame of the capacitance image, which is significantly different from that in a vertically facing contact. This serves as a typical example of projection sensing under a non-vertically facing posture.

[0083] S320. Perform time-domain and spatial-domain analysis on the capacitance image sequence corresponding to the projection area to obtain multi-dimensional features.

[0084] The projection area refers to the effective electrode coordinate area in the projected capacitive sensor array where the capacitance value changes due to the capacitive coupling effect caused by the target object entering the sensing space. This effective electrode coordinate area is determined by comparing the capacitance image sequence with the reference capacitance image and filtering out the set of coordinate points where the capacitance change exceeds a preset capacitance threshold.

[0085] The capacitance image sequence corresponding to the projection area is time-series image data that is continuously acquired in chronological order and retains only the capacitance values ​​of valid coordinate points within the projection area.

[0086] In one possible implementation, the temporal domain analysis process and the spatial domain analysis process are implemented independently and in parallel, effectively avoiding mutual interference between analysis logics of different dimensions. Specifically, the temporal domain analysis uses the time axis as the analysis benchmark. For the same coordinate point within the projection area, the capacitance value in the capacitance image sequence is extracted frame by frame. The capacitance change rate is obtained by calculating the capacitance difference between adjacent frames; the capacitance change amplitude is obtained by statistically analyzing the capacitance fluctuation amplitude over multiple consecutive frames; and the steady-state continuity feature is obtained by identifying the duration for which the capacitance value tends to stabilize, thereby capturing the dynamic behavior characteristics of the target object in the sensing space over time. The spatial domain analysis uses a single-frame capacitance image as the analysis benchmark. Based on the preset two-dimensional coordinate system of the projected capacitance sensor array, it analyzes the spatial distribution pattern, spatial clustering pattern, effective sensing area, and spatial center coordinates of the capacitance values ​​at each coordinate point within the projection area. A preset spatial clustering algorithm is used to divide the dense capacitance change areas, extracting the static spatial features of the target object on the projected capacitance sensor array, such as its spatial position and contour shape. After completing the temporal and spatial analysis, the extracted temporal dynamic features and spatial static features are integrated and classified, redundant feature data are removed, and standardization and normalization are performed to form a multi-dimensional feature that can characterize the behavior and spatial state of the target object. This multi-dimensional feature contains both dynamic change information in the temporal dimension and spatial distribution information in the spatial dimension, providing data support for the subsequent generation of control strategies for the target device.

[0087] It should be noted that since capacitance image sequences are spatiotemporally coupled data with both spatial distribution and temporal variation attributes, spatial domain analysis is responsible for mining the static spatial distribution patterns of the projected region within a single frame of capacitance image, while temporal domain analysis is responsible for mining the dynamic variation patterns of multiple frames of capacitance images over time. These two methods decompose the capacitance change information caused by the target object from different dimensions, complementing each other without feature overlap, thus forming a multi-dimensional feature set covering spatial morphology, dynamic changes, and behavioral trends. Specifically, spatial domain analysis takes the projected region in a single frame of capacitance image as the analysis object. By extracting and calculating the magnitude, distribution density, geometric contour, center coordinates, and capacitance change intensity gradient of each coordinate point within the projected region, static spatial features such as the target object's position, contact / sensing area, and spatial posture within the sensing space can be obtained, solving the problem of only sensing capacitance changes but not being able to locate the spatial state of the target object. Temporal analysis uses a sequence of multiple consecutive capacitance images as its analysis object. By analyzing and calculating the amplitude, rate of change, duration, temporal fluctuation trend, and inter-frame differences in capacitance values ​​within the projection area over time, dynamic temporal features such as the target object's movement speed, approach / remote actions, dwell time, and dynamic trajectory can be obtained. This compensates for the deficiency of single-frame spatial features in reflecting the target object's behavior process and changing trends. It combines static spatial information with dynamic temporal information to reconstruct the target object's behavioral characteristics within the sensing space by obtaining spatial dimension features through spatial analysis and temporal analysis. Therefore, by performing temporal and spatial domain analyses on the capacitance image sequence corresponding to the projection area, multi-dimensional features can be obtained.

[0088] For example, the vehicle's sensing control system first acquires multiple frames of capacitance images according to a preset fixed sequence using a projected capacitive sensor array built into the vehicle (this projected capacitive sensor array can be installed in easily accessible locations for adults, such as the vehicle's central control area, the inside of the door panel, or the side of the seat; it consists of several transmitting and receiving electrodes arranged in a matrix, with each electrode corresponding to a unique spatial coordinate, used to detect the approach and contact actions of an adult's hand or limb). Each frame of the capacitance image records the measured capacitance value of all electrode coordinate points on the projected capacitive sensor array. Then, the acquired capacitance image sequence is sequentially processed by mean filtering and median filtering. The mean filtering is used to smooth out small fluctuations in capacitance values ​​and reduce random noise interference; the median filtering is used to eliminate abnormal jumps in single-point capacitance values ​​(such as sudden capacitance changes caused by sudden electromagnetic interference) and retain valid signals. After filtering, the real-time capacitance value of each electrode coordinate point in each frame of the capacitance image is compared with the reference capacitance value pre-calibrated before the vehicle leaves the factory or during the initialization phase (i.e., the inherent capacitance value of each coordinate point of the projected capacitive sensor array under conditions of no adult approach and no interference; this reference value is pre-stored in the vehicle's storage unit and can be calibrated periodically) to obtain the capacitance change ΔC(x,y) at each coordinate point. Here, x and y represent the row and column coordinates of the projected capacitive sensor array, respectively, and the capacitance change ΔC(x,y) directly reflects the degree of capacitance change at that coordinate point caused by an adult approaching or touching it. Then, the capacitance change ΔC(x,y) is compared with a preset sensing threshold T (this preset sensing threshold T can be pre-set according to the sensitivity parameters of the vehicle's sensor array and the intensity of interference in the vehicle's interior environment, used to distinguish between adult sensing signals and environmental noise). Points with ΔC(x,y) greater than a preset sensing threshold T are marked as valid sensing points (i.e., points where capacitance changes are caused by an adult's approach or contact). Points with ΔC(x,y) less than or equal to the preset sensing threshold T are identified as invalid interference points and removed. Then, a connected component analysis algorithm is used to cluster the marked valid sensing points. Valid sensing points that are spatially continuous and exhibit consistent capacitance change trends are grouped into one class, while isolated, discrete valid points are removed (these are often caused by residual interference signals, not by adult sensing). Finally, the continuous valid sensing area formed by the clustering is determined as the projection area of ​​an adult (e.g., an adult's hand or limb) on the plane of the projected capacitive sensor array. This projection area is characterized using a minimum bounding rectangle or polygon, and its boundary coordinates and range reflect the coverage, spatial position, and contact pattern of the adult's hand or limb on the sensing plane of the projected capacitive sensor array.

[0089] Based on this, the vehicle's sensor control system further extracts multi-dimensional features for generating vehicle control strategies based on the projection area. In the spatial dimension, by calculating the centroid coordinates (X0, Y0) of the projection area (i.e., the average value of the coordinates of all effective sensing points in the projection area, used to locate the core operating position of an adult's hand or limb), the projection area S (i.e., the total number of effective sensing points in the projection area, used to characterize the contact range between the adult's hand or limb and the sensor array), and the aspect ratio k of the circumscribed rectangle (i.e., the ratio of the length to the width of the smallest circumscribed rectangle of the projection area, used to distinguish different postures of an adult's hand, such as a flat palm versus outstretched fingers), the above features can characterize the adult's spatial operating position and limb shape. In the capacitance dimension, the average capacitance change ΔC_avg (i.e., the arithmetic mean of all effective sensing points ΔC(x,y)) and the maximum capacitance change ΔC_max (i.e., the maximum value among all effective sensing points ΔC(x,y)) of all effective sensing electrode points within the projection area are extracted. These two features characterize the coupling strength between the adult's hand or limb and the sensor array. Higher coupling strength indicates closer contact or proximity, providing a quantitative basis at the capacitance level for determining operational intent. In the temporal dimension, the rate of change of the projection area (i.e., the ratio of the difference in projection area between two adjacent frames to the projection area of ​​the previous frame) and the centroid displacement velocity (i.e., the ratio of the difference in centroid coordinates between two adjacent frames to the sampling time interval) of the projection area corresponding to multiple consecutive frames of capacitance images are compared to characterize the movement trend and operational dynamics of the adult's hand or limb (e.g., the hand moving closer, further away, sliding, or remaining still). The spatial, capacitance, and temporal features calculated from the projection area together constitute the multi-dimensional features required for subsequent vehicle control strategy generation.

[0090] The control method provided in this application effectively eliminates environmental interference and invalid noise signals by performing preprocessing operations such as noise reduction, differential calculation, and connected component filtering on the capacitance image sequence. This determines the projection area of ​​the target object on the projected capacitance sensor array, thereby achieving the positioning and boundary division of the effective sensing range of the target object and avoiding interference from non-target area data in subsequent analysis. On this basis, further temporal change analysis in the time domain and spatial distribution analysis in the spatial domain are performed on the capacitance image sequence corresponding to the projection area. This extracts multi-dimensional features including the target object's position, shape, dynamic changes, and coupling strength. This ensures the relevance of multi-dimensional feature extraction and improves the reliability of multi-dimensional feature data, laying a data foundation for the subsequent generation of scientific and adaptive target device control strategies.

[0091] Optionally, the above method processes the capacitance image sequence to determine the projection area of ​​the target object on the projected capacitance sensor array, including: Based on a preset capacitance threshold, the capacitance values ​​at each coordinate point in the capacitance image sequence are thresholded to obtain a binary image sequence.

[0092] The preset capacitance threshold is a critical capacitance value that is determined and stored in the MCU's built-in memory unit through multiple calibration tests during the control system's factory manufacturing phase. This preset capacitance threshold is used to distinguish between effective capacitance changes caused by the target object and invalid interference signals caused by environmental noise and circuit drift. It can be divided into a reference capacitance value. (x,y) and the capacitance change threshold Δ .

[0093] The capacitance value at each coordinate point in the capacitance image sequence refers to the measured capacitance value C(x,y,t) collected at sampling time t for each frame of the capacitance image corresponding to the unique coordinate (x,y) of each physical electrode on the projected capacitance sensor array. The coordinates (x,y) correspond one-to-one with the row and column electrodes of the projected capacitance sensor array.

[0094] In one possible implementation, for each frame in the capacitance image sequence, all coordinate points are traversed row by row and column by column, and the current capacitance value C(x,y,t) at that coordinate point is calculated relative to the reference capacitance value using the following formula (1). The change in capacitance at (x,y) ΔC(x,y,t) = C(x,y,t) - (x,y) formula (1) Then, the calculated capacitance change ΔC(x,y,t) is compared with the preset capacitance change threshold Δ A numerical comparison is performed. If the capacitance change ΔC(x,y,t) is greater than or equal to the preset capacitance change threshold Δ If the coordinate point is determined to be a valid point for sensing the target object, its pixel value is assigned to 1; if the capacitance change ΔC(x,y,t) is less than the preset capacitance change threshold Δ If the coordinate point is determined to be an interference point without effective sensing, its pixel value is assigned to 0, thereby completing the conversion from a single-frame capacitance image to a single-frame binary image.

[0095] Finally, according to the temporal order of the capacitance image sequence, the above-mentioned threshold comparison and numerical assignment operations for each coordinate point are repeated for each frame of the image. Finally, the binary images of multiple consecutive frames after thresholding are combined in time order to form a binary image sequence corresponding to the temporal order of the original capacitance image sequence. This binary image sequence contains only two values, 0 and 1, which can remove noise interference in the original capacitance data and clearly highlight the projection area of ​​the target object in the sensor array sensing space.

[0096] Each frame of a binary image sequence is analyzed to aggregate adjacent pixels with the same pixel value into a connected region.

[0097] In this context, adjacent pixels refer to the pixels surrounding a given pixel (x, y) in the pixel matrix of a single-frame binary image that conform to a preset neighborhood rule. This application employs an industry-standard and feasible 4-neighborhood determination rule, whereby the adjacent pixels of a given pixel are the pixels directly above (x, y+1), directly below (x, y-1), directly to the left (x-1, y), and directly to the right (x+1, y). To improve the accuracy of aggregation, this can be extended to an 8-neighborhood rule (e.g., by adding adjacent pixels in the diagonal direction). The preset neighborhood rule can be set according to the detection requirements of the target device.

[0098] Having the same pixel value means that the discrete pixel values ​​of adjacent pixels are the same (both are 0 or both are 1).

[0099] A connected region is a continuous set of pixels formed by aggregating multiple adjacent pixels with the same pixel value. The boundary range of this continuous set of pixels can be characterized by parameters such as boundary pixel coordinates and the total number of pixels. Each connected region corresponds to a portion of the effective sensing area of ​​the target object on the projected capacitive sensor array.

[0100] In one possible implementation, a single-frame image from a binary image sequence is first read, and the pixel matrix information of that frame is extracted. The spatial coordinates (x, y) and pixel value (e.g., 0 or 1) of each pixel are determined, establishing a one-to-one correspondence between pixel coordinates and pixel values. Then, a secondary denoising process (e.g., morphological opening operation) is performed on the single-frame image to remove isolated stray pixels (i.e., discrete points whose pixel values ​​are different from all their neighboring pixel values) caused by sensor noise and environmental interference. This avoids stray pixels interfering with subsequent connected region aggregation. Subsequently, pixel traversal initialization is performed on the denoised single-frame image, marking all unanalyzed pixels to prepare for the identification and aggregation of subsequent neighboring pixels.

[0101] Next, from the denoised binary image of a single frame, the first unmarked target pixel (e.g., pixel value 1) is traversed and searched for. This pixel is used as a seed point and assigned a unique connected region identifier (ID). Then, starting from the seed point, all neighboring pixels of the seed point are recursively or iteratively searched according to a preset 4-neighborhood rule. It is determined whether the pixel value of the neighboring pixel is the same as that of the seed point (e.g., all are 1). If they are the same, the neighboring pixel is marked as belonging to the same connected region as the seed point, and the search for neighboring pixels of the neighboring pixel continues until no new neighboring pixels with the same pixel value can be found, thus completing the aggregation of a connected region. The above operation is repeated to continue traversing the remaining unmarked pixels in the single frame image until all valid pixels with a pixel value of 1 are marked and aggregated into the corresponding connected regions, while ignoring the aggregation of invalid pixels with a pixel value of 0. Finally, after aggregation, the parameters of each connected region are recorded. These parameters include at least: the boundary coordinates of the connected region, the total number of pixels, and the centroid coordinates. The information of all connected regions is bound to the corresponding single-frame image timing information and transmitted to the MCU, providing accurate regional data support for subsequent generation of projection regions and extraction of multi-dimensional features based on connected regions. This effectively solves the technical problem that discrete pixels cannot reflect the actual sensing range of the target object, and realizes the identification and division of the effective sensing area of ​​the target object.

[0102] From each connected region, extract the connected region with the largest area as the projection region of the target object.

[0103] Each connected region refers to a continuous set of pixels whose pixel values ​​are valid sensing identifiers, obtained by analyzing the binary image sequence frame by frame and aggregating them using the 4-neighbor or 8-neighbor rules. Each connected region has been assigned an independent region ID and has its corresponding set of pixel coordinates, boundary coordinates, and other basic information recorded synchronously.

[0104] In one possible implementation, since each connected region contains both the real sensing area generated by the coupling capacitance of the target object and a small amount of noise area that has not been filtered out after preprocessing, as well as small discrete areas formed by environmental interference, further screening is required to eliminate invalid area interference. Therefore, the area of ​​each connected region is first quantized. In this application, the total number of effective pixels contained in the connected region is used as the area representation of the connected region. The calculation method can be expressed by the following formula (2): = Formula (2) in, Let be the area of ​​the i-th connected region. Given the total number of pixels with valid sensor identifiers within the i-th connected region, the MCU sequentially traverses all marked connected regions, calculating and storing the area values ​​for each. Then, using a numerical comparison algorithm, it sorts the area values ​​of all connected regions in descending order, selecting the region with the largest area (i.e., the largest number of pixels) and locking its region ID and corresponding coordinates. Other smaller, interfering connected regions are then removed. This largest area connected region is the projection area of ​​the target object.

[0105] The control method provided in this application performs threshold processing on the capacitance values ​​of each coordinate point in the capacitance image sequence based on a preset capacitance threshold. This effectively removes environmental interference and minor noise signals, transforming the original capacitance data into a binary image sequence with clear boundaries and prominent features, thus achieving effective separation of the target sensing signal and the background signal. Furthermore, connected component analysis is performed on each frame of the binary image sequence, aggregating adjacent pixels with the same pixel value into connected regions. This allows for the localization of areas where capacitance changes caused by the target object are concentrated, avoiding recognition bias caused by discrete points. Finally, the connected region with the largest area is extracted from each connected region as the projection region of the target object. This eliminates the influence of stray interference regions, ensuring that the extracted projection region accurately reflects the actual sensing distribution of the target object. This provides a stable and reliable regional basis for subsequent feature extraction and intent recognition, improving the accuracy of subsequent feature analysis and control decisions.

[0106] Optionally, the above method performs temporal and spatial domain analysis on the capacitance image sequence corresponding to the projected region to obtain multi-dimensional features, including: Based on the temporal changes of the capacitance image sequence corresponding to the projection region, the liveness features are obtained.

[0107] The capacitance image sequence corresponding to the projection area refers to the time-series capacitance data set that is selected from the original capacitance image sequence and contains only all pixels (i.e., effective sensing pixels) within the projection area. It is not the complete original capacitance image sequence. The selection process involves comparing the boundary coordinates of the projection area with the pixel coordinates of each frame of capacitance image, extracting the real-time capacitance values ​​of all pixels whose coordinates fall within the projection area, and integrating them in chronological order to form exclusive time-series capacitance data.

[0108] Temporal variation refers to the dynamic changes in the time-series capacitance data corresponding to the projected area over time. It includes at least the amplitude, period, frequency, and fluctuation pattern of capacitance value changes. It is used to reflect the dynamic changes in the coupling relationship between the target object and the sensor array. It is also the key to distinguishing between living and non-living objects, because living objects have physiological dynamic changes that cause periodic and regular fluctuations in capacitance value; non-living objects do not have such dynamic changes, and their capacitance value tends to be stable.

[0109] Liveness characteristics refer to dynamic capacitance changes based on physiological activities that can uniquely characterize a target object as a living organism, distinguishing it from the static capacitance characteristics of non-living organisms (such as plastic imitation fingers or metal blocks). These characteristics include at least: capacitance fluctuation period, fluctuation amplitude threshold, and dynamic response speed.

[0110] In one possible implementation, the capacitance values ​​of all pixels belonging to the projection region in each frame of the capacitance image are first extracted through coordinate filtering. The average capacitance value of the projection region (i.e., the arithmetic mean of the capacitance values ​​of all pixels in the projection region in each frame) is calculated. The average capacitance values ​​of multiple consecutive frames are sorted by timestamp to form a time-domain capacitance change curve corresponding to the projection region. The horizontal axis of this time-domain capacitance change curve is time, and the vertical axis is the average capacitance value of the projection region, recording the trend of capacitance value change over time. Then, the time-domain capacitance change curve is standardized. First, a low-pass filtering algorithm is used to remove high-frequency noise caused by environmental interference and sensor noise, retaining the effective capacitance change signal. Then, the rate and amplitude of capacitance value change are calculated through differential operation. The periodicity of capacitance fluctuation is extracted through a periodic detection algorithm (such as the autocorrelation function method). Dynamic change parameters that conform to the physiological characteristics of living organisms (such as fluctuation period, maximum fluctuation amplitude, fluctuation stability, etc.) are selected, while static or irregular fluctuation signals corresponding to non-living organisms are removed. Finally, a preset living feature rule library is called, and the dynamic change parameters are matched one by one with the living feature thresholds in the preset living feature rule library. If the dynamically changing parameters meet the criteria for determining liveness features in the preset liveness feature rule base (e.g., fluctuation period and amplitude are both within the preset threshold range, and the fluctuation pattern is continuous and stable), the target object is determined to be live. Simultaneously, the dynamically changing parameters are integrated and quantified to obtain the corresponding liveness features (e.g., "capacitance fluctuation period 1.2s, fluctuation amplitude 8%"). If the criteria are not met, the target object is determined to be non-live, and no liveness features are generated. The preset liveness feature rule base pre-stores the time-domain change feature thresholds corresponding to live objects (e.g., the capacitance fluctuation period caused by blood circulation in the fingertip of a human finger is 0.5-2s, and the fluctuation amplitude is 5%-15% of the preset baseline capacitance value), as well as the time-domain change features of non-live objects (e.g., capacitance value fluctuation amplitude is less than 3%, no fixed period).

[0111] It should be noted that when a living object (such as a human finger or palm) is coupled to a projected capacitive sensor array, its physiological activities (such as blood circulation, slight muscle contractions, and micro-movements of the limbs caused by respiration) will cause small, periodic changes in the contact area and coupling distance with the sensor electrodes, which in turn will cause periodic, regular fluctuations in the capacitance value of the projected area. In contrast, when a non-living object (such as a plastic finger or metal sheet) is coupled to the sensor, its coupling state remains fixed, and the capacitance value of the projected area tends to be stable in the time domain, with no obvious fluctuations or irregular fluctuations. Therefore, by analyzing the time domain changes of the capacitance image sequence corresponding to the projected area, dynamic features unique to living objects can be extracted, thus distinguishing between living and non-living objects. This solves the technical defects of traditional capacitive sensing technology, which cannot identify living objects and is easily interfered with by non-living object spoofing operations.

[0112] Spatial contour features are calculated based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image.

[0113] Among them, multi-dimensional features include: liveness features and spatial contour features.

[0114] Among them, a single-frame capacitance image refers to any one time-series image in the capacitance image sequence, which records the real-time capacitance values ​​of all electrode coordinate points on the sensor array at that moment. Each pixel (corresponding to the electrode coordinate) corresponds to a unique capacitance value, forming a spatial distribution matrix of capacitance values.

[0115] The spatial distribution of capacitance value refers to the distribution pattern of the capacitance value of all pixels (i.e., electrode coordinates) in the projection area according to their spatial position (x, y). That is, the difference and trend of capacitance value at different spatial positions. This distribution pattern is determined by the coupling distance and contact area between the target object and each electrode of the sensor array (e.g., the central area of ​​the target object is more coupled with the sensor electrode, resulting in a larger capacitance value, while the edge area is less coupled, resulting in a smaller capacitance value).

[0116] Spatial contour features refer to a set of features that can characterize the spatial shape, boundary features, and location distribution of a projection area. They are a component of multi-dimensional features and are used to reflect the spatial posture and sensing range of the target object.

[0117] In one possible implementation, based on the projection area, a single-frame capacitance image corresponding to that projection area is retrieved, and the capacitance value data of all pixels within the projection area is extracted to construct a spatial distribution matrix C(x,y) of the capacitance values ​​in the projection area. Here, x and y are the spatial coordinates of the pixels within the projection area, and C(x,y) is the real-time capacitance value corresponding to that coordinate point. The spatial distribution matrix is ​​then preprocessed, and gradient features of capacitance value changes are extracted using gradient calculation (such as the Sobel operator). This locates the boundary pixels of the projection area, i.e., pixels whose capacitance values ​​abruptly change from high to low. The lines connecting these boundary pixels constitute the actual contour of the projection area. Then, based on the coordinate data of the boundary pixels, quantization parameters of the spatial contour features are calculated. These quantization parameters include at least: the set of contour boundary coordinates, the centroid coordinates of the projection area, the parameters of the contour's circumscribed rectangle, and the contour complexity.

[0118] The contour boundary coordinate set records the (x, y) coordinates of all boundary pixels in the projection area, representing the specific shape of the contour. The centroid coordinates of the projection area are obtained by calculating the weighted average of the coordinates of all pixels within the projection area (where the weight is the capacitance value of the corresponding pixel, with a higher capacitance value resulting in a higher weight), representing the sensing position of the target object. The parameters of the contour circumscribed rectangle include at least the length L, width W, and aspect ratio k = W / L, representing the overall shape of the projection area (e.g., elongated, circular, etc.). The contour complexity is obtained by calculating the curvature changes of the boundary pixels and the degree of contour convexity, representing the edge details of the target object.

[0119] Then, all the quantization parameters obtained from the above calculations are integrated to form standardized spatial contour features.

[0120] The control method provided in this application, based on the temporal changes of the capacitance image sequence corresponding to the projection area, can analyze the living features used to distinguish between real living objects and non-living interference objects, thereby eliminating interference factors such as spurious objects and static foreign objects from the temporal dimension and improving the realism and anti-interference ability of target object detection. At the same time, based on the spatial distribution of capacitance values ​​in a single frame capacitance image of the projection area, spatial contour features characterizing the shape, size and spatial distribution of the target object can be calculated, providing a reliable static quantitative basis for the shape determination and region positioning of the target object. The multi-dimensional features in this application can be composed of living features and spatial contour features. Through the organic combination of temporal dynamic features and spatial static features, a multi-dimensional and three-dimensional feature characterization of the target object can be achieved.

[0121] Optionally, multidimensional features include: liveness features.

[0122] Among them, living characteristics are the signal characteristics of the periodic physiological activities of living organisms.

[0123] The above method analyzes the liveness features based on the temporal changes of the capacitance image sequence corresponding to the projection region, including: Signal extraction is performed on the capacitance image sequence corresponding to the projection area to obtain the time-domain signal of capacitance change in the projection area.

[0124] Here, capacitance change refers to the change in real-time capacitance value at each coordinate point within the projection area compared to the pre-calibrated reference capacitance value of the sensor array. It is not the real-time capacitance value itself, which refers to the inherent capacitance value at each coordinate point of the sensor array when no target object enters the sensing space. (x,y), this value has been calibrated and stored in the device's storage unit before the target device leaves the factory, and is used as the basis for calculating the capacitance change.

[0125] In one possible implementation, the calibrated projection area coordinate range (x1, y1) to (x2, y2) is first retrieved, along with the corresponding capacitance image sequence, establishing a one-to-one correspondence between the projection area coordinates and the pixel coordinates in the capacitance image sequence. Then, in chronological order, single-frame capacitance images from the capacitance image sequence are read frame by frame. For each frame, only the real-time capacitance value C(x, y, t) of all pixels (i.e., effective sensing electrode points) within the projection area coordinate range (x∈[x1, x2], y∈[y1, y2]) is extracted, where t is the timestamp corresponding to the single frame image. Capacitance values ​​of all coordinate points outside the projection area are simultaneously ignored to avoid interference from invalid signals. Next, the extracted projection area capacitance values ​​from each frame undergo preliminary purification to remove abnormal capacitance values ​​caused by sensor noise and environmental interference, ensuring the stability of the extracted capacitance signal. Finally, the extracted and purified projection area capacitance values ​​are organized in chronological order according to their timestamps to form a projection area capacitance signal set synchronized with the capacitance image sequence.

[0126] After completing the signal extraction operation, first retrieve the reference capacitance values ​​of each coordinate point in the preset projection area. (x,y), and simultaneously retrieve the real-time capacitance signal set C(x,y,t) of the projection area; then, for the timestamp t corresponding to each frame of capacitance image, calculate the capacitance change ΔC(x,y,t) of each coordinate point (x,y) in the projection area using the above formula (1), so as to convert the real-time capacitance value into a capacitance change that can reflect the sensing intensity of the target object. If ΔC(x,y,t) is positive, it indicates that the coupling strength between the target object and the electrode at that coordinate point is enhanced (e.g., closer); if it is negative, it indicates that the coupling strength is weakened (e.g., farther away); if it is 0, it indicates that there is no effective sensing. After that, the capacitance change ΔC(x,y,t) of all coordinate points in the projection area is averaged to obtain the average capacitance change ΔC_avg(t) of the projection area corresponding to each timestamp t, which can be expressed by the following formula (3): ΔC_avg(t) = [ΣΔC(x,y,t)] ÷ (M×N) Formula (3) Where M represents the number of pixels in the projection area along the x-axis, and N represents the number of pixels in the projection area along the y-axis. Through averaging, the discrete capacitance change signals at multiple coordinate points within the projection area are integrated into a unified signal that characterizes the overall sensing state of the entire projection area, avoiding interference from fluctuations in individual coordinate point signals. Subsequently, the average capacitance change ΔC_avg(t) corresponding to all timestamps t is arranged chronologically to form a time-series signal with time t as the horizontal axis and the average capacitance change ΔC_avg(t) as the vertical axis. This time-series signal is the time-domain signal of the capacitance change in the projection area. Finally, the time-domain signal is standardized to determine its temporal resolution (consistent with the sampling frequency of the capacitance image sequence) and amplitude range, eliminating abrupt changes in the time sequence to ensure the continuity and stability of the time-domain signal.

[0127] Spectral analysis is performed on the time-domain signal to extract the periodic fluctuation characteristics of the time-domain signal within the preset vital sign frequency band; the periodic fluctuation characteristics are used as living characteristics.

[0128] The preset vital sign frequency band refers to a specific frequency range stored in the device rule base that corresponds to human vital signs (such as heart rate and respiration). After extensive experimental calibration, this frequency band is specifically set to 0.1Hz to 3Hz, where 0.1Hz to 0.5Hz corresponds to the human respiratory frequency range and 0.5Hz to 3Hz corresponds to the human heart rate frequency range. This frequency band setting can effectively eliminate the influence of environmental interference (such as high-frequency electromagnetic noise and low-frequency mechanical vibration) and ensure that the extracted features are only related to human vital signs.

[0129] In one possible implementation, the time-domain signal is first processed to remove the DC component, eliminating static offsets. Then, a preset low-pass filter algorithm is used to filter high-frequency interference signals (e.g., frequencies above 10Hz, since the frequency band for human vital signs is below 10Hz), resulting in a denoised time-domain signal. Subsequently, an FFT (Fast Fourier Transform) is performed on the denoised time-domain signal to convert it into a frequency-domain signal, yielding the amplitude-frequency response curve, A(f), where f is the frequency variable and A(f) is the signal amplitude at the corresponding frequency f. The FFT decomposes the continuous time-domain signal into a superposition of several sine waves of different frequencies, calculating the amplitude of each frequency component to clarify the frequency distribution characteristics of the time-domain signal. Next, based on the frequency-domain signal, signal components with frequencies f falling within a preset frequency band of 0.1Hz to 3Hz are selected, and all interference components outside this band are eliminated. Finally, periodic analysis is performed on the selected frequency band signal to extract its periodic fluctuation characteristics. For example, the peak frequency f0 (corresponding to heart rate or respiratory rate), peak amplitude A0 (corresponding to the capacitance change amplitude caused by vital signs), periodic stability parameter (i.e., the coefficient of variation of the time interval between two adjacent signal peaks, used to characterize the stability of vital signs), and harmonic distribution characteristics of the signal (the harmonic amplitude of live vital sign signals decays regularly, while non-live signals do not have this characteristic) are all periodic fluctuation characteristics caused by the vital signs of the target object. They can reflect the unique physiological activity patterns of living organisms and provide core basis for subsequent liveness determination.

[0130] Finally, the extracted periodic fluctuation features are validated to determine whether they meet preset validity thresholds (e.g., peak amplitude A0 must be greater than a preset amplitude threshold to ensure sufficient signal strength; periodic stability parameter must be less than a preset coefficient of variation threshold to ensure stability of the fluctuation; peak frequency f0 must fall within a preset vital sign frequency band of 0.1Hz to 3Hz). If the preset validity thresholds are met, the periodic fluctuation feature is determined to be a valid feature and subsequently defined as a liveness feature. If the preset validity thresholds are not met, the periodic fluctuation feature is determined to be an invalid feature (likely caused by interference signals) and is not considered a liveness feature; the process must return to the previous steps for signal acquisition and processing.

[0131] The control method provided in this application includes multi-dimensional features, including living features, which are signal features characterizing the periodic physiological activities of a living organism. In the feature extraction process, the capacitance image sequence corresponding to the projection area is first analyzed and extracted to obtain a time-domain signal that reflects the capacitance change pattern within the projection area. Then, the time-domain signal is subjected to spectrum analysis to extract the periodic fluctuation features of the time-domain signal within the preset vital sign frequency band. These periodic fluctuation features are directly used as living features. Through this feature extraction method based on capacitance time-series changes and spectrum analysis, the unique physiological activity signals of living organisms can be captured, and living targets can be reliably distinguished from non-living interference objects. This avoids the problem of misidentification caused by non-living objects triggering sensor signals, while improving the stability and recognizability of the extracted living features.

[0132] Optionally, spatial contour features include geometric features.

[0133] Among them, geometric features refer to the shape, size, and positional distribution of the projected area of ​​the target object on the projected capacitive sensor array.

[0134] The above method calculates spatial contour features based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image, including: Based on the single-frame capacitance image corresponding to the projection area, the area of ​​the projection area and the coordinates of the centroid of the projection area on the projected capacitance sensor array are calculated; the area and coordinates are used as geometric features.

[0135] In one possible implementation, a single-frame capacitance image corresponding to the current projection area to be processed is retrieved from the capacitance image sequence. Then, the pixel matrix information of the single-frame capacitance image is read to determine the pixel value (such as 0 or 1) and pixel coordinates (x, y) of each pixel. The mapping relationship between the pixel coordinates (x, y) and the physical coordinates (X, Y) of the projected capacitance sensor array electrode is established through preset coordinate calibration parameters, that is, the conversion is achieved through the following formula (4).

[0136] X = x × kx + X0, Formula (4): Y = y × ky + Y0 Where kx and ky are the scaling factors of pixel coordinates and sensor array physical coordinates in the X and Y axes, respectively (which can be pre-calibrated by the electrode spacing and image resolution of the sensor array and stored in the device storage unit), and X0 and Y0 are the sensor array physical coordinate offsets corresponding to the upper left pixel of a single frame of capacitive image.

[0137] In a single-frame capacitive image, all pixels corresponding to the projection area are located and marked. The pixel coordinates (x, y) and corresponding pixel values ​​(all 1) of all pixels within this area are recorded. Invalid pixels (with a pixel value of 0) outside the projection area are removed, forming a pixel dataset specific to the projection area. Based on this pixel dataset specific to the projection area, the area of ​​the projection area and the coordinates of the centroid of the projection area on the projected capacitive sensor array are calculated.

[0138] The process of calculating the area of ​​the projected region is as follows: First, based on the pixel dataset dedicated to the projection area, traverse all pixels in the pixel dataset and count the total number of valid pixels with a pixel value of 1, denoted as N (N is a positive integer and N≥1; if N=0, the projection area is determined to be invalid, and the connected component analysis step is re-executed). Then, obtain the physical area parameter S0 corresponding to a single pixel that is pre-stored in the device storage unit. This physical area parameter S0 is determined by the electrode spacing of the projected capacitive sensor array and can be expressed by the following formula (5): S0 = d_x × d_y formula (5) Where d_x is the spacing between adjacent electrodes in the X-axis direction of the sensor array, and d_y is the spacing between adjacent electrodes in the Y-axis direction (unit: mm). The physical area parameter S0 is a pre-calibrated fixed value that can be flexibly adjusted according to the specifications of the sensor array.

[0139] The actual physical area S of the projected area is then calculated using the following formula (6).

[0140] S = N × S0 = N × dx × dy formula (6) The actual physical area S is measured in mm². It is calculated by multiplying the number of pixels by the physical area of ​​a single pixel, thus representing the size of the target object's projection coverage on the sensor array. Finally, after calculation, the projection area S is bound to the corresponding single-frame image temporal information and the projection area identifier ID for subsequent comprehensive analysis of multi-dimensional features.

[0141] The process of calculating the coordinates of the centroid of the projected region on the projected capacitive sensor array is as follows: First, read the pixel coordinates (x_i, y_i) of all valid pixels in the projection area (i=1,2,...,N, where N is the total number of valid pixels). At the same time, read the capacitance change ΔC_i corresponding to each pixel (this data comes from the original capacitance data after preprocessing the capacitance image sequence; ΔC_i is the difference between the real-time capacitance value and the reference capacitance value corresponding to pixel (x_i, y_i), reflecting the coupling strength between the target object and the electrode corresponding to the pixel). Then, using the capacitance change ΔC_i of each pixel as a weighting coefficient, calculate the centroid coordinates (x_c, y_c) of the projection area at the pixel level of the single-frame capacitance image using the following formula (7).

[0142] X_c= , Y_c= Formula (7) If the difference in capacitive coupling strength does not need to be considered, an equal-weighted averaging algorithm can be used (i.e., all ΔC_i=1). In this case, the above formula (7) can be simplified to formula (8): X_c= , Y_c= Formula (8) The two algorithms of formula (7) and formula (8) above can be selected according to the detection requirements of the target equipment.

[0143] Then, the pixel-level centroid coordinates (x_c, y_c) are converted into the physical centroid coordinates (X_c, Y_c) on the projected capacitive sensor array using the coordinate mapping relationship of the above formula (4), that is: X_c = x_c × k_x + X0, Y_c = y_c × k_y + Y0. Among them, k_x, k_y, X0, and Y0 are consistent with the coordinate mapping parameters mentioned above, ensuring that the converted physical coordinates accurately correspond to the actual electrode positions of the sensor array.

[0144] Once the calculation is complete, the physical coordinates of the centroid (X_c, Y_c) are bound to the projected area S, the projected area identifier ID, and the temporal information of the single-frame image for storage. These centroid coordinates reflect the center position of the target object within the sensor array's sensing space, providing core spatial feature basis for subsequent target object posture recognition, operation intent parsing, and control strategy generation.

[0145] Therefore, the area S of the projected region is used to quantitatively characterize the coverage and size of the region on the sensing plane corresponding to the projected capacitive sensor array, directly reflecting the spatial scale attribute of the projected area of ​​the target object. The centroid coordinates (X_c, Y_c) of the projected region are used to characterize the spatial location of the region in the two-dimensional coordinate system of the projected capacitive sensor array, directly reflecting the spatial positioning attribute of the projected area of ​​the target object. The spatial scale information reflected by the area and the spatial position information reflected by the centroid coordinates are core parameters for quantitatively describing the projected region from the perspective of spatial geometry and spatial position. They do not involve non-geometric attributes such as capacitance change, temporal change, and signal amplitude. They only objectively depict the geometric shape, spatial distribution, and positional relationship of the projected region. Therefore, the area and centroid coordinates of the projected region are uniformly classified as geometric features that can intuitively reflect the spatial geometric characteristics of the target object. These geometric features can stably and intuitively reflect the projection shape and positional distribution of the target object on the sensing plane.

[0146] The control method provided in this application includes spatial contour features, which are geometric features. These geometric features are the shape, size, and positional distribution characteristics of the projected area of ​​the target object on the projected capacitive sensor array. They can intuitively depict the spatial occupancy and orientation attributes of the target object from a spatial dimension, providing basic spatial feature support for subsequent target type discrimination and attitude recognition. Based on the single-frame capacitive image corresponding to the projected area, the area of ​​the projected area and the coordinates of the center of gravity of the projected area on the projected capacitive sensor array are calculated. The area and coordinates are used together as geometric features. The area can quantify the size of the projected coverage of the target object, effectively distinguishing target objects of different sizes. The center of gravity coordinates can locate the actual position of the target object in the sensing space, intuitively reflecting its spatial orientation distribution. The combination of the two can construct a spatial geometric representation of the target object, improving the reliability and accuracy of subsequent multi-dimensional feature analysis and target perception.

[0147] Optionally, spatial contour features may also include statistical features.

[0148] Among them, statistical characteristics are statistical parameters obtained by calculating the capacitance value within the projected area of ​​the target object.

[0149] The above method calculates spatial contour features based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image, including: Based on the single-frame capacitance image corresponding to the projection area, the mean and variance of the capacitance values ​​within the projection area are calculated, and the mean and variance are used as statistical features.

[0150] In one possible implementation, based on the single-frame capacitance image corresponding to the projection area, the capacitance change ΔC_i corresponding to all valid pixels in its pixel dataset is retrieved. ΔC_i is the difference between the real-time capacitance value of the sensor electrode corresponding to each pixel and the baseline capacitance value in the absence of a target, which is a numerical value characterizing the coupling strength between the target object and the electrode. Subsequently, an arithmetic mean algorithm is used to calculate the mean, which can quantify the average level of the overall capacitance coupling strength in the projection area. It can intuitively distinguish the differences in the overall coupling degree between different targets such as human bodies, child safety seats, and non-biological objects and the sensor array, providing a basic statistical basis for the preliminary classification of target objects.

[0151] After obtaining the mean capacitance value, based on the same set of effective capacitance change ΔC_i data, the dispersion characteristic of the capacitance value, i.e., the variance σ², is further calculated. The calculation formula is: σ²=Σ[(ΔC_i)] i [μ)²] / N. This variance measures the fluctuation range and uniformity of the capacitance change of each pixel relative to the mean. The smaller the variance value, the more uniform the distribution of capacitive coupling strength caused by the target object, corresponding to targets with regular shapes and uniform dielectric properties. The larger the variance value, the more obvious the difference in capacitive coupling strength distribution, corresponding to targets with irregular shapes and uneven dielectric properties.

[0152] Therefore, it can be seen that the mean, which reflects the overall average level of capacitance change within the projected area, is a classic statistic characterizing the central tendency of data and can reflect the overall capacitive coupling strength between the target object and the sensor array; the variance, which reflects the fluctuation range and distribution uniformity of each capacitance change relative to the mean, is a classic statistic characterizing the degree of data dispersion and can reflect the spatial distribution differences of the capacitive response. Both are obtained solely from the mathematical statistical calculation of capacitance values ​​and do not involve other dimensional attributes such as the target's spatial shape, coordinate position, texture details, and temporal micro-motions. They are clearly demarcated from geometric features, texture features, and liveness features. At the same time, they can distinguish the capacitive response characteristics of different targets from a numerical statistical perspective, providing supplementary discrimination criteria for target recognition beyond geometric features. Therefore, the mean and variance are combined into statistical features.

[0153] The control method provided in this application further includes statistical features in its spatial contour features. These statistical features are statistical parameters calculated based on the capacitance values ​​within the projection area of ​​the target object. They can characterize the coupling characteristics between the target object and the projected capacitive sensor array from the dimension of capacitance value distribution, effectively supplementing the discrimination dimension that geometric features can only characterize spatial morphology. Based on the single-frame capacitance image corresponding to the projection area, the mean and variance of the capacitance values ​​within the projection area are calculated, and the mean and variance are used as statistical features. The mean can quantitatively characterize the overall average level of capacitance coupling strength within the projection area, intuitively reflecting the overall coupling strength between the target object and the sensor array. The variance can characterize the dispersion and distribution uniformity of the capacitance values ​​relative to the mean, reflecting the spatial fluctuation differences in capacitance response. The combination of the two can provide a stable statistical basis for target object type differentiation and posture recognition.

[0154] Optionally, spatial contour features may also include: texture features.

[0155] Among them, texture features are the image texture patterns of capacitance values ​​within the projection area of ​​the target object.

[0156] The above method calculates spatial contour features based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image, including: Calculate the corresponding gray-level co-occurrence matrix based on the single-frame capacitance image corresponding to the projection region.

[0157] The gray-level co-occurrence matrix is ​​a joint probability distribution matrix of pixel gray-level values ​​that appear in pairs along a preset direction and interval, based on the gray levels of the capacitance image. The preset direction refers to the spatial arrangement angle of pixel pairs, typically including four directions: horizontal, vertical, and diagonal; the preset pixel interval is the pixel distance between paired pixels.

[0158] In one possible implementation, the calculation parameters of the gray-level co-occurrence matrix are first set. The total number of gray levels after normalization and quantization of capacitance change is denoted as G (usually 8, 16, or 32 levels, which can be calibrated according to the detection accuracy). The preset pixel interval d is set (d is a positive integer, generally 1, representing adjacent pixels), and four typical calculation directions are determined, namely 0° (horizontal to the right), 45° (upper right diagonal), 90° (vertical upward), and 135° (upper left diagonal). The pixel coordinate offset rules corresponding to each direction are: 0° direction is (x+d, y), 45° direction is (x+d, yd), 90° direction is (x, yd), and 135° direction is (xd, yd). Then, the entire effective pixel in the projection area of ​​the single-frame capacitive image is traversed row by row and column by column. The effective pixel (x, y) traversed at the current time is used as the reference pixel. According to the selected calculation direction and the preset pixel interval d, the corresponding paired pixel (x', y') is searched. Only when the paired pixel is also within the effective pixel range of the projection area is the pixel pair determined to be an effective pixel pair. The gray values ​​a and b of the reference pixel and the paired pixel are read respectively, where a and b are integers ∈ [0, G-1]. For each effective pixel pair with a gray value of (a, b) identified, the frequency value f(a, b) at the position of the a-th row and b-th column in the frequency statistics matrix of the G×G dimension is incremented by 1. After traversing all effective pixels, the original frequency matrix f(a, b) of the gray value pixel pair is obtained, where a and b are the gray levels of the reference pixel and the paired pixel, respectively, and f(a, b) is the total number of times the gray value combination appears. The frequency matrix is ​​then normalized, which can be expressed by the following formula (9): P (a, b)=f(a, b) ÷ N_total formula (9) Where N_total is the sum of the total frequencies of all valid pixel pairs in the projection area, and P(a, b) is the joint probability of gray-level pair (a, b), with a value range of [0,1]. After converting all frequency values ​​into probability values ​​through formula (9), the gray-level co-occurrence matrix P(a, b) with G×G dimensions and perfectly matching the current projection area is finally obtained.

[0159] Based on the gray-level co-occurrence matrix, contrast and energy parameters are extracted; these parameters are then used as texture features.

[0160] The contrast ratio represents the degree of local grayscale variation in the capacitance image, reflecting the magnitude of the difference in grayscale values ​​of pixels at different locations within the projection area and the intensity of local changes. It characterizes the magnitude of local fluctuations in the capacitive coupling strength induced by the target object within the sensing space. A large contrast ratio indicates significant differences in grayscale values ​​within the projection area, corresponding to irregular shapes or uneven dielectric distribution of the target object. A small contrast ratio indicates smooth grayscale changes in the capacitance values, corresponding to regular shapes and uniform dielectric properties of the target object, allowing for direct differentiation of local texture differences between different targets. The energy parameter represents the degree of grayscale distribution in the capacitance image, characterizing the concentration and uniformity of probability values ​​in the grayscale co-occurrence matrix. It reflects the regularity of the grayscale distribution within the projection area. A larger energy parameter indicates a more concentrated grayscale distribution and a simpler, more uniform texture structure; a smaller energy parameter indicates a more dispersed grayscale distribution and a more complex and varied texture structure.

[0161] In one possible implementation, after the gray-level co-occurrence matrix corresponding to the projection area is constructed, based on the joint probability distribution of each gray-level value pair within the matrix, the classic gray-level co-occurrence matrix feature statistics method is used to calculate the two core texture feature parameters, contrast and energy, respectively. The contrast can be calculated using the following formula (10).

[0162] Contrast= Formula (10) The energy parameters can be calculated using the following formula (11).

[0163] Energy= Formula (11) In the above formulas (10) and (11), G is the total number of gray levels preset for the capacitance image. These are the row and column indices of the gray-level co-occurrence matrix, respectively. The corresponding pixel's gray value is represented by P(a, b), which is the normalized joint probability of the corresponding position in the gray-level co-occurrence matrix. By traversing all rows and columns of the gray-level co-occurrence matrix and performing the above formula calculation, the unique contrast and energy parameters corresponding to the current projection area can be obtained.

[0164] Since contrast and energy parameters are both feature parameters that are directly calculated based on the gray-level co-occurrence matrix and used to quantify the texture distribution law of capacitance images, they only reflect the spatial variation and distribution characteristics of capacitance gray values ​​within the projection area. They are completely different from the spatial morphology of geometric features and the numerical central tendency dimension of statistical features. Therefore, the two are integrated as texture features that characterize the texture properties of the target object in this application.

[0165] The control method provided in this application further includes texture features in the spatial contour features. These texture features are used to characterize the image texture pattern presented by the capacitance value in the projection area of ​​the target object. They can characterize the dielectric properties and surface morphology differences of the target object from the dimension of the texture distribution of the capacitance response, effectively making up for the limitations of geometric features in describing only spatial morphology and statistical features in describing only numerical distribution, and supplementing the target perception with a new texture dimension. Based on the single-frame capacitance image corresponding to the projection area, the corresponding gray-level co-occurrence matrix is ​​calculated. This gray-level co-occurrence matrix is ​​a matrix formed by statistically analyzing the joint probability distribution of pixel gray values ​​that appear in pairs under preset directions and preset intervals, using the gray levels of the capacitance image as the statistical basis. It serves as the carrier for quantifying the texture pattern of the capacitance image. Then, contrast and energy parameters are extracted based on the gray-level co-occurrence matrix. The contrast parameter is used to characterize the degree of local gray-level change in the capacitance image, intuitively reflecting the fluctuation difference of the capacitance coupling strength and the roughness of the texture within the projection area. The energy parameter is used to characterize the regularity of the gray-level distribution of the capacitance image, corresponding to the uniformity of the target dielectric properties and the concentration of the texture distribution. Using the contrast parameter and energy parameter together as texture features can capture the specificity of different target objects in the capacitance response texture pattern, further enriching the representation dimension of spatial contour features.

[0166] Figure 4 A flowchart illustrating a control method provided in an embodiment of this application. Figure 3 .like Figure 4 As shown, the above method generates a control strategy for the target device based on multi-dimensional features, including: S410. Based on multi-dimensional features, determine the attribute categories of the target object.

[0167] In one possible implementation, the features of each dimension are first preprocessed by dimensional normalization, and the linear normalization formula of the following formula (12) is used to uniformly map the feature data of different magnitudes.

[0168] = Formula (12) in, These are the original eigenvalues. and These are the preset minimum and maximum calibration thresholds for this type of feature, respectively. The normalized feature values ​​are used to eliminate classification bias caused by differences in the numerical range of area, coordinates, capacitance, and texture parameters. Then, all normalized features are used to construct a one-dimensional feature vector, and the attribute category is determined by a preset classification decision rule. This preset classification decision rule is either a threshold discrimination logic formed by combining a large number of sample calibrations or a lightweight embedded classification model. Specifically, the threshold discrimination method configures reasonable value ranges for each feature for different target object attribute categories (such as adult occupants, child occupants, child safety seats, non-biological objects, and empty state). For example, if the projected area is larger than the adult calibration threshold and the average coupling strength is within the human characteristic range, it is identified as an adult occupant; if the area is within the child range and there are live micro-movement features, it is identified as a child occupant; if the area matches a child seat but there are no live features and the texture uniformity matches the seat's dielectric properties, it is identified as a child safety seat; if the area is small and there are no live features, it is identified as an ordinary non-biological object; if the number of effective pixels is 0, it is identified as an empty object. If a lightweight classification model is used, the feature vector is input into a decision tree trained offline, and the decision tree outputs the confidence probability of each category. The category with the highest confidence is selected as the final judgment result. The final determined target object attribute category can clearly distinguish between live occupants and non-live objects.

[0169] S420. If the attribute category of the target object belongs to the protection category, then the spatial pose of the target object is determined based on multi-dimensional features.

[0170] In one possible implementation, if the target object's attribute category belongs to the protection category, multi-dimensional features are input into a preset spatial attitude determination model. This preset spatial attitude determination model can employ a feature threshold-based rule matching model or a lightweight machine learning classification model, and is pre-trained and calibrated using a large number of samples and stored in the device's storage unit. The preset spatial attitude determination model uses the offset of the center of gravity relative to the center of the sensor array, the degree of deformation of the projected area, the fluctuation characteristics of the mean and variance of capacitance, and the variation law of texture parameters as the determination criteria to classify the spatial attitude of the target object into key attitude types such as normal sitting, leaning forward, sitting sideways, lying down, and limbs extended. By comparing the feature values ​​with the preset attitude determination rules, a unique spatial attitude determination result for the current target object is output.

[0171] S430. Generate a control strategy for the target device based on the attribute category or spatial attitude of the target object.

[0172] In one possible implementation, the target object's attribute category is identified based on a comprehensive analysis of geometric, statistical, textural, and liveness features. This category is then matched with pre-defined classification decision rules to generate a differentiated control command. The target object's attribute category includes at least: adult occupant, child occupant, child safety seat, non-biological object, and unoccupied state. For example, when the identification result is an adult occupant, it is determined to be a valid occupant state, and a control strategy is generated that allows the vehicle's airbags to ignite and deploy normally. When the identification result is a child occupant, a tiered control strategy is generated to reduce the intensity of the vehicle's airbag deployment or delay deployment. When the identification result is a child safety seat, non-biological object, or unoccupied state, a suppression control strategy is generated to prohibit the deployment of the vehicle's airbags, preventing safety risks caused by malfunctioning safety devices. This attribute-based control strategy generation method can ensure the rationality and safety of the control strategy from the perspective of the target's essential attributes.

[0173] Based on the determined target object attribute category, the spatial posture information obtained by combining geometric features such as the centroid coordinates and coordinate offset of the projection area is further analyzed to execute control strategy correction and adjustment. The spatial posture of the target object includes at least dangerous or unconventional postures such as normal sitting posture, leaning forward, side sitting, and deviation from the center of the seat. For example, when an adult occupant is identified as being in a dangerous posture leaning forward close to the dashboard, correction commands for adjusting the airbag ignition timing and optimizing deployment pressure are added to the normal ignition strategy; when an occupant is identified as being in a side sitting posture, a selective deployment strategy for the corresponding side airbags is generated to adapt to the occupant protection needs under different postures and maximize the safety protection effect.

[0174] The control method provided in this application, based on multi-dimensional features, determines the attribute category of the target object, distinguishing different types of target objects such as human occupants, child safety seats, and non-biological objects, thus achieving reliable classification and identification of the target object. If the attribute category of the target object is determined to belong to a protection category requiring safety protection, the spatial posture of the target object is further determined by multi-dimensional feature analysis. This can capture key spatial posture information such as occupant leaning forward, sitting sideways, and positional deviation, effectively compensating for the limitations of single-dimensional feature recognition. Finally, by combining the attribute category and spatial posture of the target object, an adaptive control strategy for the target device matching the current detection scenario is generated. This strategy can not only adjust the execution logic of the safety device according to the differences in the target category to avoid false triggering or missed triggering, but also match the optimal safety protection scheme for different spatial postures, significantly improving the adaptability and overall reliability of the target device safety control.

[0175] Optionally, multidimensional features include: liveness features.

[0176] Among them, living characteristics are the signal characteristics of the periodic physiological activities of living organisms.

[0177] The above method determines the attribute categories of the target object based on multi-dimensional features, including: Based on living characteristics, determine whether the target object possesses life characteristics.

[0178] In one possible implementation, by performing high-pass filtering, time-domain smoothing, and periodic extraction on the capacitance change at the same electrode coordinate point in a time-series capacitance image, the regular micro-motion capacitance signal generated by the target object's vital activities such as breathing and heartbeat can be obtained. This liveness feature can distinguish the periodic physiological fluctuations of living organisms from the static or irregular disturbances of non-living organisms. Specifically, the processing unit built into the target device compares the extracted liveness feature, which includes at least: micro-motion frequency, fluctuation amplitude, and periodic regularity, with a pre-calibrated and stored life characteristic judgment threshold. If the liveness feature falls within the preset physiological micro-motion threshold range, the target object is determined to have vital characteristics; otherwise, it is determined not to have vital characteristics.

[0179] If the object exhibits life characteristics, its attribute category is determined to be an organism.

[0180] In one possible implementation, when the target object is determined to have periodic micro-motion signals that conform to human physiological laws by comparing live features, i.e., confirming that it has life characteristics, the target object is directly labeled as a living organism according to the preset target attribute classification rules.

[0181] If it does not possess life characteristics, then the attribute category of the target object is determined to be a non-living entity.

[0182] In one possible implementation, when it is determined that the target object has no periodic physiological micro-movements, only static capacitance distribution or irregular disturbance signals, that is, when it is confirmed that the target object does not have life characteristics, its attribute category is determined as a non-living body according to the preset classification logic.

[0183] The control method provided in this application includes multi-dimensional features, including living features. These living features are signal characteristics that characterize the periodic physiological activities of living organisms. They can capture the periodic changes in capacitance caused by micro-movements such as breathing and heartbeat in the human body from the time domain dimension, effectively overcoming the technical limitations of geometric features, statistical features, and texture features, which can only reflect spatial information and cannot distinguish life attributes. Based on these living features, it can be determined whether the target object has life characteristics. If the target object is determined to have life characteristics, its attribute category is determined to be a living organism. If the target object is determined not to have life characteristics, its attribute category is determined to be a non-living organism. This enables rapid initial classification of the life attributes of the target object and effectively eliminates interference from non-biological life-like organisms such as humanoid dolls and robots.

[0184] Optionally, the multi-dimensional features in the above methods also include geometric features.

[0185] Among them, geometric features refer to the shape, size, and positional distribution of the projected area of ​​the target object on the projected capacitive sensor array.

[0186] The above method, which determines the attribute category of the target object based on multi-dimensional features, also includes: If the target object's attribute category is "organism", then based on the area and coordinates in the geometric features, determine whether the area is greater than or equal to a preset area threshold.

[0187] The preset area threshold is a critical area value obtained through extensive experimental calibration. This preset area threshold is determined by the difference in the normal projected area of ​​adults and children on the seat and has been stored in the storage unit of the target device.

[0188] In one possible implementation, if the target object's attribute category is "organism", the calculated projection area is retrieved, and through numerical comparison, it is directly determined whether the area is greater than or equal to a preset area threshold, thereby completing the first layer of screening for adult and child classification.

[0189] If the area is greater than or equal to a preset area threshold, the organism's attribute category is determined to be adult.

[0190] In one possible implementation, when the projected area is greater than or equal to a preset area threshold, the attribute category of the organism can be directly determined as an adult, taking into account the physiological characteristics that adult occupants have a larger projection coverage area and a larger physical footprint than children on vehicle seats. This determination logic is based on the physical size differences of real human body shapes, and the preset area threshold has fully considered the minimum projected area of ​​adults of different heights and weights, eliminating the possibility of misjudgment for smaller adults.

[0191] If the area is less than the preset area threshold and the coordinates meet the preset coordinate conditions, then the attribute category of the organism is determined to be a child.

[0192] The preset coordinate conditions are the normal center of gravity coordinate range of a child occupant on a vehicle seat, which is calibrated in advance through experiments. The normal center of gravity coordinate range includes at least: the lateral centering range in the X-axis direction, the longitudinal sitting range in the Y-axis direction, and the offset limit of the center of gravity coordinate relative to the center of the sensor array. The preset coordinate conditions have been stored in the storage unit of the target device and can match the normal sitting posture spatial position characteristics of the child occupant.

[0193] In one possible implementation, when the projected area is smaller than a preset area threshold, it only indicates that the projected size of the target organism is too small. To avoid misjudgment caused by the organism's sitting posture, curling up, or positional shift, a secondary verification is needed by combining the physical coordinates of the center of gravity in the geometric features, i.e., determining whether the coordinates meet the preset coordinate conditions. Only when the area is smaller than the preset area threshold and the coordinates meet the preset coordinate conditions is the organism's attribute category finally determined to be a child. By combining the two conditions of area and coordinates, misclassification problems caused by small-area interference factors such as small animals or partial limb occlusion can be effectively eliminated.

[0194] The control method provided in this application includes geometric features in its multi-dimensional features. These geometric features characterize the shape, size, and positional distribution of the projection area of ​​the target object on the projected capacitive sensor array. They can depict the spatial occupancy and orientation information of the target object from a spatial dimension, providing a reliable basis for the refined subtype classification of organisms. Given that the target object's attribute category is determined to be an organism, the method uses the area and coordinate information in the geometric features to determine whether the area is greater than or equal to a preset area threshold. If the area is greater than or equal to the preset area threshold, the attribute category of the organism is directly determined to be an adult. This allows for efficient differentiation between adults and other organisms of different body types based on the difference in the projected body size, ensuring real-time classification and recognition. If the area is less than the preset area threshold and the coordinates meet preset coordinate conditions, the attribute category of the organism is determined to be a child. This dual-judgment logic, combining initial screening with preset area thresholds and verification with preset coordinate conditions, effectively avoids misjudgments caused by crouching posture, positional shifts, or local interference, improving the accuracy of identifying adult and child organism subtypes.

[0195] Optionally, multidimensional features may also include statistical features.

[0196] Among them, statistical characteristics are statistical parameters obtained by calculating the capacitance value within the projected area of ​​the target object.

[0197] The above method, which determines the attribute category of the target object based on multi-dimensional features, also includes: If the target object's attribute category is a living organism, then based on the mean and variance in the statistical features, it is determined whether the mean is greater than a preset mean threshold and the variance is less than a preset variance threshold.

[0198] Among them, the preset mean threshold and the preset variance threshold are critical values ​​obtained by experimental calibration of the coupling characteristics between the human body (such as adults and children) and the sensor array in a large number of real vehicle scenarios. They have fully considered the statistical differences in capacitive response of adults of different body sizes and children of different ages, and are stored in the storage unit of the target device.

[0199] In one possible implementation, once the target object's attribute category is determined to be a living organism, the calculated mean and variance of the projection area capacitance are automatically retrieved. A dual-condition joint verification is then performed through numerical comparison: simultaneously determining whether both conditions are met, namely, whether the mean is greater than a preset mean threshold and whether the variance is less than a preset variance threshold. This serves as the statistical basis for classifying adults and children. This dual-condition judgment logic can effectively avoid misjudgments caused by the influence of posture and sitting position on a single geometric feature (such as area), thereby improving classification accuracy.

[0200] If the mean is greater than the preset mean threshold and the variance is less than the preset variance threshold, then the attribute category of the organism is determined to be adult.

[0201] In one possible implementation, if the mean is greater than a preset mean threshold and the variance is less than a preset variance threshold, the attribute category of the organism can be directly determined as adult by combining the physical coupling characteristics of the adult occupant and the projected capacitive sensor array. From a technical perspective, adult occupants have a larger body size, a wider contact coverage with the sensor array, and more uniform dielectric properties of human skin and muscles, resulting in a higher overall capacitive coupling strength (corresponding to a larger mean) in the projection area. At the same time, the capacitance change of each electrode pixel fluctuates less relative to the mean (corresponding to a smaller variance). The preset mean threshold and preset variance threshold sufficiently cover the differences in body size and dielectric properties of most adults, ensuring that the judgment logic will not misjudge due to the smaller body size or slight deviation in sitting posture of adults.

[0202] If the mean is less than or equal to a preset mean threshold and the variance is greater than or equal to a preset variance threshold, then the attribute category of the organism is determined to be child.

[0203] In one possible implementation, if the mean is less than or equal to a preset mean threshold and the variance is greater than or equal to a preset variance threshold, the biological attribute category can be ultimately determined to be a child by combining the physical coupling characteristics of the child occupant and the projected capacitive sensor array. From a technical perspective, because child occupants are much smaller than adults, their contact coverage with the projected capacitive sensor array is smaller, resulting in lower overall capacitive coupling strength (corresponding to a smaller mean). Furthermore, the dielectric properties of a child's body are relatively more unevenly distributed, and their sitting posture is more prone to slight shifts, leading to more significant fluctuations in the capacitance of each electrode pixel relative to the mean (corresponding to a larger variance). The preset mean threshold and preset variance threshold have fully considered the child's body size and dielectric characteristics. Combined with preliminary verification of geometric features, interference factors from incomplete human bodies such as small animals and parts of a child's body can be effectively eliminated.

[0204] The control method provided in this application also includes statistical features in its multi-dimensional features. These statistical features are statistical parameters calculated based on the capacitance values ​​within the projected region of the target object. They can characterize the differences in coupling characteristics of the target object from the perspectives of the central tendency and discrete distribution of the capacitance response, effectively compensating for the limitations of single geometric features in biological subdivision scenarios and improving the anti-interference ability of adult and child classification. Based on the target object's attribute category already determined to be a biological object, a joint conditional judgment is performed based on the mean and variance in the statistical features, i.e., determining whether the mean is greater than a preset mean threshold and the variance is less than a preset variance threshold. If the above conditions are met, the attribute category of the organism is determined to be adult, taking into account the higher capacitive coupling strength and more uniform and stable capacitance value distribution of adults with the projected capacitive sensor array. If the mean is less than or equal to a preset mean threshold and the variance is greater than or equal to a preset variance threshold, the attribute category of the organism is determined to be child, taking into account the relatively lower capacitive coupling strength and higher dispersion of capacitance value distribution of children with the sensor array. By jointly judging the dual statistical parameters of mean and variance, the classification interference caused by differences in limb posture and sitting posture can be reduced, and stable distinction between adult and child organisms can be achieved.

[0205] Optionally, multi-dimensional features also include: texture features.

[0206] Among them, texture features are the image texture patterns of capacitance values ​​within the projection area of ​​the target object.

[0207] The above method, which determines the attribute category of the target object based on multi-dimensional features, also includes: If the target object's attribute category is a living organism, then based on the contrast and energy parameters in the texture features, it is determined whether the contrast is less than a preset contrast threshold and whether the energy parameter is greater than a preset energy threshold.

[0208] Wherein, contrast represents the degree of local grayscale change in the capacitor image; energy parameter represents the degree of grayscale distribution in the capacitor image.

[0209] Among them, the preset contrast threshold and preset energy threshold are critical discrimination values ​​obtained by calibration through a large number of measured samples of adult and child occupants, and have been pre-fixed in the discrimination rule library of the processing unit.

[0210] In one possible implementation, after the target object has been determined to belong to the organism category through comprehensive multi-dimensional feature analysis, the process of distinguishing between adults and children based on texture features is initiated. First, the calculation results of contrast and energy parameters are retrieved, and numerical comparison operations are performed separately. Simultaneously, it is determined whether the contrast is less than a preset contrast threshold and whether the energy parameter is greater than a preset energy threshold. Through the joint logic judgment of the two parameters, the texture dimension is used as the basis for subsequent organism subtype classification.

[0211] If the contrast is less than the preset contrast threshold and the energy parameter is greater than the preset energy threshold, then the organism's attribute category is determined to be adult.

[0212] In one possible implementation, if the contrast is less than a preset contrast threshold and the energy parameter is greater than a preset energy threshold, the textural characteristics of an adult occupant interacting with the projected capacitive sensor array are considered. Specifically, the adult body has a large coverage area and uniform surface dielectric properties, resulting in smooth local grayscale changes and low contrast in the capacitive image, while the capacitance values ​​are concentrated and well-distributed, leading to higher energy parameters. Based on this texture feature matching rule, the current organism's attribute category is ultimately determined to be "adult," and the target object is assigned an "adult" classification label. Simultaneously, the classification result is bound and stored with the corresponding texture feature parameters and temporal information.

[0213] If the contrast is greater than or equal to a preset contrast threshold and the energy parameter is less than or equal to a preset energy threshold, then the organism's attribute category is determined to be a child.

[0214] In one possible implementation, if the contrast is greater than or equal to a preset contrast threshold and the energy parameter is less than or equal to a preset energy threshold, and considering the texture characteristics of child occupants compared to adults—namely, the smaller body size, relatively larger range of limb movements, and irregular and discrete distribution of the capacitive coupling area with the projected capacitive sensor array—that is, the capacitive image corresponding to a child has drastic local grayscale changes, higher contrast, dispersed capacitance values, low texture regularity, and consequently lower energy parameters, the attribute category of the organism is determined to be a child based on the texture feature discrimination rules, and the child classification label is marked and data is stored.

[0215] The control method provided in this application includes multi-dimensional features, including texture features. These texture features characterize the image texture pattern corresponding to the capacitance value within the projection area of ​​the target object. This supplements the classification criteria for organisms from the dimension of the texture distribution of the capacitance response, overcoming the limitations of geometric and statistical features, and further improving the accuracy and anti-interference ability of adult and child classification. Contrast is used to reflect the degree of local grayscale change in the capacitance image, and the energy parameter characterizes the regularity of the grayscale distribution in the capacitance image. Based on the target object's attribute category being determined as an organism, a joint conditional judgment is performed based on the contrast and energy parameter in the texture features, i.e., determining whether the contrast is less than a predetermined value. If a contrast threshold is set and the energy parameter is greater than a preset energy threshold, and the above conditions are met, then the attribute category of the organism is determined to be an adult, based on the characteristics that the grayscale change of the capacitance response generated by the interaction between the adult and the sensor array is gradual and the grayscale distribution is more concentrated and uniform. If the contrast is greater than or equal to the preset contrast threshold and the energy parameter is less than or equal to the preset energy threshold, then the attribute category of the organism is determined to be a child, based on the characteristics that the grayscale fluctuation of the capacitance response is more significant and the grayscale distribution is more discrete. Through the collaborative discrimination of contrast and energy dual texture parameters, the classification error caused by sitting posture changes, limb occlusion and other scenarios can be effectively reduced, and the refined identification of organism subtypes can be achieved.

[0216] Optionally, before determining the spatial pose of the target object based on multi-dimensional features, the above method further includes: The attribute categories of the target object are determined based on a preset category protection mapping table.

[0217] Among them, the preset category protection mapping table is a key-value mapping data table that is pre-built and stored in the built-in MCU storage unit of the target device before the target device leaves the factory or during the installation and commissioning stage, through a large number of real vehicle experiments and scenario calibrations. The table has preset target object attribute category entries with clear classifications, including at least: adults, children, child safety seats, non-biological objects, and no-load status, and at the same time, each attribute category is bound with a corresponding protection attribute tag.

[0218] In one possible implementation, after identifying the attribute category of the target object through geometric features, statistical features, texture features, and time-domain features, and obtaining a clear classification result such as living organism (e.g., adult, child) or non-living organism, the control system retrieves the preset category protection mapping table, and sequentially performs matching retrieval between the identified target object attribute category and all category entries pre-stored in the preset category protection mapping table. By using character matching or index value comparison, the corresponding entry of the current target object attribute category in the preset category protection mapping table is locked, and the attribute category lookup confirmation is completed.

[0219] If the attribute category belongs to a category in the preset category protection mapping table, then the attribute category is determined to belong to the protection category.

[0220] In one possible implementation, after matching the target object's attribute category with the preset category protection mapping table, if the identified attribute category matches the corresponding entry in the preset category protection mapping table that is marked as a protected object, such as adults, children, or other biological categories that require collision protection from vehicle safety devices, then according to the preset safety control logic, the attribute category of the target object is directly determined to belong to the protection category, and a protection category marker is generated simultaneously. This marker is then bound and stored with the target object's attribute category, geometric features, statistical features, and other data.

[0221] If the attribute category does not belong to a category in the preset category protection mapping table, then the attribute category is determined not to belong to a protection category.

[0222] In one possible implementation, if the attribute category of the current target object does not match the entry marked as a protected object in the preset category protection mapping table after checking the preset category protection mapping table, that is, the corresponding categories such as child safety seats, non-biological objects, and unloaded objects that do not require the protection of safety devices, then the attribute category is confirmed not to belong to the protection category according to the preset judgment rules, and a non-protection category mark is generated and written to the temporary storage area of ​​the target device.

[0223] The control method provided in this application matches and determines the attribute category of the target object based on a preset category protection mapping table. If the attribute category is determined to belong to a protected object category pre-recorded in the preset category protection mapping table, then the attribute category is clearly identified as a protection category, which can quickly identify the target type that requires targeted security protection for the target device. If the attribute category is determined not to be included in the preset category protection mapping table, then the attribute category is determined not to belong to a protection category, and there is no need to trigger the corresponding security protection mechanism. Through the standardized mapping table matching method, the determination logic of category attribution can be simplified, the identification efficiency can be improved, and the identification error caused by manual or fuzzy determination can be effectively avoided.

[0224] Optionally, if the attribute category of the target object belongs to the protection category in the above method, the spatial pose of the target object is determined based on multi-dimensional features, including: If the attribute category of the target object belongs to the protection category, then the shape parameters of the projected area of ​​the target object on the projected capacitive sensor array are calculated based on the geometric features in the multi-dimensional features.

[0225] Among them, geometric features are the comprehensive spatial characteristics of the projected area of ​​the target object within the physical space corresponding to the projected capacitive sensor array, including its shape contour, size, and spatial orientation. These features are the sole data basis for calculating shape parameters.

[0226] In one possible implementation, when the target object's attribute category belongs to the protection category, it is determined that the target object may be a biological entity requiring protection by the passive safety system of the target device, such as an adult occupant or a child occupant, thus improving system processing efficiency and reducing unnecessary calculations. After identifying the target object in the protection category, the extracted geometric features from the multi-dimensional features are retrieved. Then, based on the pixel contour data of the projection area corresponding to these geometric features, the shape parameters of the target object's projection area are calculated using a preset contour extraction algorithm in digital image processing. These shape parameters include at least quantitative indicators such as the aspect ratio, roundness, and contour complexity of the projection area.

[0227] The specific calculation process is as follows: In the determined binary projection area, the area with a pixel value of 1 is the effective area of ​​the target object, and the area with a pixel value of 0 is the background area. For example, the 8-neighborhood serial boundary tracking algorithm commonly used in embedded platforms is used to traverse the boundary pixels. First, the first effective pixel point located at the edge is found in the projection area as the starting pixel point. Then, the 8 neighboring pixels of the current pixel point are detected in a fixed counterclockwise direction (such as top, bottom, left, right, upper left, upper right, lower left, lower right). Only edge pixels with a pixel value of 1 that are adjacent to the background pixels are selected. The pixel coordinates (x_i, y_i) of each boundary pixel are tracked and recorded in turn. Redundant pixels without contour meaning inside the projection area are removed throughout the process, and finally an ordered and closed set of boundary pixel coordinates {(x1,y1), (x2,y2), …, (x_M,y_M)} is formed, where M is the total number of boundary pixels. After obtaining the discrete boundary pixel coordinate set, a preset polygon approximation algorithm is used to fit the contour of the discrete points to retain the key features of the projection area shape while reducing computational redundancy. Specifically, the process involves using the first and last points of the boundary pixel set as the baseline line segment, calculating the vertical distance from each of the remaining boundary points in the boundary pixel coordinate set to the baseline line segment, identifying points with a distance less than a preset contour fitting threshold as redundant points and removing them, and retaining only key feature points such as corner points and extreme points that play a decisive role in the contour shape. The above approximation operation is repeated for the segmented line segments until all remaining points meet the distance requirements. Finally, the retained key feature points are connected sequentially to form a closed polygonal contour curve that can accurately represent the shape of the target projection area. Finally, based on the closed polygon contour curve fitted above, shape parameters are extracted through basic numerical operations: First, the extreme coordinates of the contour curve in the horizontal and vertical directions are extracted, the minimum bounding rectangle of the contour is calculated, and the length L and width W of the rectangle are obtained. The aspect ratio R = L / W is further calculated to characterize whether the projected area is elongated, square, or flat. Second, the perimeter Z of the contour is obtained by accumulating the Euclidean distances between adjacent feature points of the contour curve. Third, the equivalent area A enclosed by the closed contour is calculated using a preset polygon shoelace formula. Fourth, the roundness C = 4πA / P² is calculated based on the perimeter and area. The closer the roundness is to 1, the closer the contour is to a circle; conversely, the lower the roundness, the more irregular the shape. All the above operations are basic arithmetic operations such as addition, subtraction, multiplication, division, squaring, and square root extraction, without complex iterations or floating-point bottlenecks. The calculation results can be directly used as shape parameters for subsequent attitude determination.

[0228] Based on the coordinates and shape parameters of geometric features, the spatial pose type of the target object is determined.

[0229] The coordinates of the geometric features are the coordinates of the centroid of the projected area on the projected capacitive sensor array. That is, after coordinate mapping transformation, the coordinates of the centroid of the projected area of ​​the target object are the physical centroid coordinates of the physical electrode coordinate system of the projected capacitive sensor array. These coordinates reflect the actual center position of the target object in the sensing space of the sensor array and are the basis for determining the position of attitude deviation.

[0230] In one possible implementation, the physical center of gravity coordinates are first fused with calculated aspect ratios, roundness, and other shape parameters. A preset posture determination rule library is then invoked. Subsequently, the actually calculated center of gravity coordinates and shape parameters are matched and compared one by one with the standard thresholds in the preset posture determination rule library. Based on the changes in coordinate offset direction, magnitude, and contour shape, the current spatial posture type of the target object is determined. For example, when the center of gravity coordinates extend forward beyond a preset range and the aspect ratio of the projected area changes accordingly, it is determined to be a forward-leaning posture; when the center of gravity coordinates shift to the side and the contour shape is asymmetrical, it is determined to be a side-sitting posture. The preset posture determination rule library predefines the coordinate ranges and shape parameter threshold ranges corresponding to various spatial posture types, such as normal sitting posture, forward-leaning posture, side-sitting posture, and center offset posture.

[0231] The control method provided in this application, if the attribute category of the target object belongs to the protection category requiring security protection, can calculate the shape parameters of the projection area of ​​the target object on the projected capacitive sensor array based on the geometric features in the multi-dimensional features. The geometric features are used to characterize the shape, size, and position distribution of the projection area of ​​the target object on the projected capacitive sensor array, which can characterize the projection contour and spatial occupancy characteristics of the target object from the spatial dimension, providing a stable morphological data foundation for attitude recognition. Furthermore, based on the coordinates of the center of gravity of the projection area on the projected capacitive sensor array in the geometric features and the obtained shape parameters, the spatial attitude type of the target object is comprehensively determined. The center of gravity coordinates reflect the spatial position and offset state of the target object, and the shape parameters quantify the contour deformation characteristics of the projection area. The collaborative discrimination of the two can improve the accuracy and anti-interference of spatial attitude determination and avoid the recognition deviation caused by a single feature.

[0232] Optionally, the method described above determines the spatial pose type of the target object based on the coordinates and shape parameters of geometric features, including: If the coordinate value of the first axis in the coordinate system of the geometric feature is greater than the first attitude threshold, then the spatial attitude type of the target object is determined to be a forward tilt attitude.

[0233] Among them, the coordinates of the geometric features are the coordinates of the center of gravity of the projected area of ​​the target object in the physical coordinate system of the projected capacitive sensor array; the first axis coordinate value is the longitudinal Y-axis coordinate value, which directly reflects the front and rear position distribution of the center of gravity of the target object in the longitudinal direction of the sensor array. The larger the first axis coordinate value, the closer the center of gravity of the target object is to the front end of the projected capacitive sensor array, that is, the closer it is to the direction of the vehicle dashboard.

[0234] The first posture threshold is a longitudinal coordinate critical value obtained in advance through a large number of occupant posture experiments. This first posture threshold is the upper limit of the longitudinal coordinate of the occupant's center of gravity in a standard sitting posture, and can be selected according to the actual situation.

[0235] In one possible implementation, the first axis coordinate value of the center of gravity is compared with a first attitude threshold. When the first axis coordinate value is determined to be greater than the first attitude threshold, the spatial attitude type of the target object is directly determined to be a forward tilting attitude, based on the physical attitude law that the center of gravity will shift forward when the occupant leans forward and the longitudinal coordinate exceeds the normal range.

[0236] If the difference between the shape parameter and the reference orientation angle is greater than the second attitude threshold, the spatial attitude type of the target object is determined to be a side-sitting attitude.

[0237] The shape parameters include at least the contour orientation angle of the target object's projection area. This contour orientation angle is obtained by fitting the minimum bounding rectangle to the boundary pixels of the projection area and calculating the principal axis of the contour, and is used to characterize the overall orientation angle of the projection area. The reference orientation angle is the baseline orientation angle corresponding to the target object's projection area under a standard sitting posture. It is a fixed angle value obtained through numerous standard sitting posture experiments and can be selected according to the actual situation.

[0238] The second posture threshold is the critical value for determining the angle deviation of the side-sitting posture, used to filter out minor angle fluctuations in normal sitting posture. It can be selected according to the actual situation.

[0239] In one possible implementation, the absolute difference between the actual orientation angle and the reference orientation angle in the shape parameters is first calculated. Then, the absolute difference is compared with a second attitude threshold. When the absolute difference is greater than the second attitude threshold, it indicates that the orientation of the target object's projection area is significantly deviated from the standard sitting posture. Combined with the morphological feature that the projection area contour direction will be significantly deviated when the occupant is sitting sideways, the spatial posture type of the target object is determined to be a side-sitting posture. This determination method is based on contour geometric angle calculation, which can effectively identify the occupant's side-sitting posture and make up for the deficiency that single coordinate determination cannot cover posture deviation scenarios.

[0240] If neither of the criteria for forward-leaning posture nor side-sitting posture is met, then the spatial posture type of the target object is determined to be the standard posture.

[0241] In one possible implementation, when the first axis coordinate value of the center of gravity is not greater than the first attitude threshold, and the difference between the shape parameter orientation angle and the reference orientation angle is not greater than the second attitude threshold (i.e., the conditions for both forward-leaning and side-sitting attitudes are not met), it indicates that the center of gravity position and the orientation of the projected area contour of the target object are within the preset normal range, conforming to the spatial position and morphological characteristics corresponding to the standard occupant sitting posture. In this case, the spatial attitude type of the target object can be directly determined as the standard attitude. This judgment rule can eliminate abnormal attitude interference, clarify the benchmark attitude for conventional safety protection, and provide a basis for subsequent airbag deployment control strategies.

[0242] The control method provided in this application, when determining the spatial posture type of a target object, if the first axis coordinate value of the centroid coordinate in the geometric features is greater than a preset first posture threshold, then the spatial posture type of the target object is determined to be a forward-leaning posture based on the coordinate offset feature. This can quickly identify the dangerous posture of the occupant's body leaning forward, providing timely judgment basis for the timing and intensity adjustment of subsequent safety devices. If the difference between the shape parameter of the projection area and the preset reference direction angle is greater than a preset second posture threshold, then the spatial posture type of the target object is determined to be a side-sitting posture based on the contour angle offset feature. This can capture the abnormal state of the occupant's side-sitting posture, make up for the blind spot of single coordinate judgment, and improve the completeness of posture recognition. If neither the determination conditions for forward-leaning posture nor side-sitting posture are met, then the spatial posture type of the target object is determined to be a standard posture. Through the dual-layer joint determination logic of coordinate threshold and shape angle difference, various typical occupant postures can be distinguished, reducing the probability of posture misjudgment and improving the reliability and robustness of spatial posture recognition.

[0243] Optionally, the method described above generates a control strategy for the target device based on the target object's attribute category or spatial pose, including: If the target object's attribute category is non-living, then an inhibition command for the target device is generated.

[0244] Among them, the suppression command is used to suppress the protective actions of the safety devices in the target equipment.

[0245] The safety devices in the target equipment can be selected according to the actual situation. For example, if the target equipment is a vehicle, the safety device can be selected as an airbag.

[0246] In one possible implementation, after determining the attribute category of the target object, if it is determined to be a non-living object such as luggage, a toy, or a child seat, since such objects do not require safety devices for protection, the preset non-living object control rules are directly triggered. Specifically, the MCU built into the target device retrieves pre-programmed suppression instruction configuration parameters from its storage unit and automatically generates the corresponding suppression instruction. This suppression instruction can be a digital level signal or a bus communication instruction.

[0247] For example, the function of the suppression command of the safety device (airbag) in the target device is to transmit control signals to the airbag control unit to prohibit ignition and deployment, block the execution path of the safety device's protective action, and avoid the airbag from accidentally deploying and causing damage to the target device or unnecessary safety hazards.

[0248] If the target object's attribute category is child, then the first control command for the target device is generated.

[0249] The first control command is used to trigger the target device to be controlled by the first intensity parameter.

[0250] The first strength parameter is lower than the calibrated strength parameter. Taking an airbag as an example, this first strength parameter is a set of control parameters such as the airbag's ignition energy, deployment pressure, and inflation rate. The calibrated strength parameter is the standard calibrated strength parameter set by the factory of the target equipment. It can be selected according to the actual situation.

[0251] In one possible implementation, when the target object is determined to be a child—a biological entity requiring flexible protection—based on multi-dimensional features, the target device, according to a preset child protection control strategy, retrieves a first strength parameter from its storage unit, obtained through real-vehicle calibration and safety testing. Since this first strength parameter is lower than the calibrated strength parameter, it is adapted to the child's physiological tolerance, avoiding secondary injury caused by high-intensity deployment. Therefore, a standardized first control command is generated based on this first strength parameter and sent to the safety device's control unit via a communication bus. This first control command triggers the safety device to execute protective actions according to the low-intensity first control parameter, ensuring the safety of the child occupant.

[0252] If the target object's attribute category is adult and its corresponding spatial pose is standard pose, then a second control command for the target device is generated.

[0253] The second control command is used to trigger the target device to calibrate the intensity parameter control.

[0254] In one possible implementation, when the target object's attribute category is determined to be an adult, and its spatial posture is determined to be a standard posture with the seat centered and the torso upright based on geometric features and shape parameters, it means that the adult occupant is in a normal and safe riding posture, which meets the applicable scenario for the standard protection of the safety device. At this time, the standard strength parameters calibrated by the vehicle manufacturer in the storage unit are called. These calibrated strength parameters are the optimal control parameters set for collision protection in the normal posture of an adult, which can provide sufficient safety protection for the adult. Based on these calibrated strength parameters, a second control command is generated and transmitted to the airbag control unit, triggering the safety device to perform normal protective actions according to the default calibrated strength parameters.

[0255] If the target object's attribute category is adult and its corresponding spatial posture is forward tilt, then a third control command for the target device is generated.

[0256] The third control command is used to trigger the target device to control the calibrated intensity parameters and adjust the control direction.

[0257] In one possible implementation, when the target object's attribute category is adult, but it is determined by its center of gravity coordinates and shape parameters to be in a dangerous spatial posture with its body leaning forward and close to the dashboard, the MCU executes control rules specific to the forward-leaning posture. On the one hand, it still uses the calibration strength parameters adapted to adult protection to ensure basic protection capabilities during a collision. On the other hand, it retrieves pre-stored control direction adjustment parameters to adaptively correct control direction-related parameters such as the airbag's deployment direction, inflation angle, and timing phase. The MCU integrates the calibration strength control logic and the direction adjustment logic to generate a third control command. This third control command, while maintaining the calibration protection strength, controls the airbag to avoid the dangerous forward-leaning position of the occupant, adjusts the airbag deployment path and direction of action, reduces the risk of rigid contact between the airbag and the occupant in the forward-leaning posture, and improves the occupant protection effect in dangerous postures.

[0258] The control method provided in this application automatically generates a suppression command for the target device when the target object is determined to be a non-living object. This suppression command restricts and blocks the protective triggering and execution of the safety device within the target device, avoiding the problem of accidental activation or triggering of the safety device in unoccupied scenarios and preventing unnecessary safety hazards and equipment damage. When the target object is determined to be a child, a first control command corresponding to the target device is generated. This first control command drives the target device to complete safety control according to a first intensity parameter. This first intensity parameter is pre-set to be lower than the standard calibration intensity parameter, which can adapt to the physical tolerance characteristics of children, reduce the operating intensity of the safety device, and prevent high-intensity protective actions. This prevents secondary injuries to children. When the target is an adult and the target is in a standard spatial posture, a second control command is generated to drive the safety device to work stably with standard calibration strength parameters, ensuring that the protective performance meets the standards under normal sitting conditions and providing balanced and reliable standard safety protection for adult occupants. When the target is an adult and in a dangerous forward-leaning posture, a third control command is generated. While maintaining the calibration strength parameters unchanged and ensuring basic protection capabilities, the control direction of the safety device is adjusted simultaneously to specifically adapt to the special forward-leaning driving posture, making up for the limitations of protection in a fixed protection direction and improving the protection accuracy, adaptability and operational safety of the passive safety system of the target equipment in complex driving scenarios.

[0259] It should be noted that, based on the above control method, this application also provides a self-calibration, fault tolerance, and safety redundancy system for the control system to ensure the accuracy and stability of occupant classification detection and the reliable operation of the target equipment's passive safety system. Specifically, in the automatic calibration phase, each time the target equipment (such as a vehicle) is started and the passenger door is closed, the control system automatically executes a background signal learning process to update the reference capacitance value. (x, y) is used to adapt to the slow drift of the capacitive signal caused by factors such as temperature changes in the central control screen and surface decorations, ensuring the accuracy of the projected capacitive sensing data. When the system detects that an occupant has left (i.e., the capacitive image returns to the background state) and the seatbelt is unfastened, the control system will use the projected capacitive sensing data of this session to fine-tune the user habit model, adapting it to the body characteristics of frequently seated occupants, and further optimizing the classification accuracy. In the fault diagnosis and degradation process, the control system will periodically perform self-check operations, such as verifying whether the capacitive sensing channel of the projected capacitive sensor array is normal by injecting test signals; if the central control screen malfunctions or is completely covered by a large object, causing the control system to be unable to provide an effective occupant classification signal, the airbag control unit will automatically revert to the traditional control strategy, that is, mainly relying on the seatbelt status and collision intensity to perform safety control, and may send a "Occupant classification system malfunction, please check" prompt message to the instrument panel to avoid safety hazards caused by control system failure and ensure the continuous effective operation of the safety system.

[0260] Based on the same inventive concept, this application also provides a control device. Since the principle of the device in this application is similar to the control method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0261] Figure 5 This is a schematic diagram of a control device provided in an embodiment of this application. Figure 5 As shown, the control device 500 may include: The acquisition module 501 is used to acquire a corresponding capacitance image sequence in response to the target object entering the sensing space of the projected capacitance sensor array of the target device. The capacitance image sequence includes multiple images of the capacitance value of each coordinate point on the projected capacitance sensor array changing over time due to the target object. The parsing module 502 is used to parse multi-dimensional features from the capacitance image sequence. The multi-dimensional features are features parsed from the time domain and spatial domain of the capacitance image sequence. The generation module 503 is used to generate a control strategy for the target device based on multi-dimensional features.

[0262] In one optional implementation, the parsing module 502 is specifically used to: process the capacitance image sequence to determine the projection area of ​​the target object on the projected capacitance sensor array; and perform temporal and spatial domain parsing on the capacitance image sequence corresponding to the projection area to obtain multi-dimensional features.

[0263] In one optional implementation, the parsing module 502 is specifically used to: perform threshold processing on the capacitance values ​​of each coordinate point in the capacitance image sequence based on a preset capacitance threshold to obtain a binary image sequence; analyze each frame of the binary image sequence to aggregate adjacent pixels with the same pixel value in each frame into a connected region; and extract the connected region with the largest area from each connected region as the projection region of the target object.

[0264] In one optional implementation, the parsing module 502 is specifically used to: parse the liveness features based on the temporal changes of the capacitance image sequence corresponding to the projection region; calculate the spatial contour features based on the spatial distribution of capacitance values ​​in a single frame capacitance image of the projection region; the multi-dimensional features include: liveness features and spatial contour features.

[0265] In one optional implementation, the multi-dimensional features include: live features, which are signal features of the periodic physiological activities of a living organism; the parsing module 502 is specifically used to: extract signals from the capacitance image sequence corresponding to the projection area to obtain the time-domain signal of capacitance change in the projection area; perform spectral analysis on the time-domain signal to extract the periodic fluctuation features of the time-domain signal within a preset vital sign frequency band; and use the periodic fluctuation features as live features.

[0266] In one optional implementation, the spatial contour features include: geometric features, which are the shape, size, and positional distribution characteristics of the projection area of ​​the target object on the projected capacitive sensor array; and the parsing module 502, which is specifically used to: calculate the area of ​​the projection area and the coordinates of the centroid of the projection area on the projected capacitive sensor array based on the single-frame capacitive image corresponding to the projection area; and use the area and coordinates as geometric features.

[0267] In an optional implementation, the spatial contour features further include: statistical features, which are statistical parameters calculated based on the capacitance values ​​within the projection area of ​​the target object; and a parsing module 502, specifically used to: calculate the mean and variance of the capacitance values ​​within the projection area based on a single-frame capacitance image corresponding to the projection area; and use the mean and variance as statistical features.

[0268] In an optional implementation, the spatial contour features further include: texture features, which are the image texture patterns of capacitance values ​​within the projection area of ​​the target object; the parsing module 502 is specifically used to: calculate the corresponding gray-level co-occurrence matrix based on the single-frame capacitance image corresponding to the projection area; the gray-level co-occurrence matrix is ​​a joint probability distribution matrix of pixel gray-level values ​​that appear in pairs in a preset direction and interval, based on the gray-level of the capacitance image; extract contrast and energy parameters based on the gray-level co-occurrence matrix, where contrast is the degree of local gray-level change in the capacitance image; and energy parameters are the degree of gray-level distribution in the capacitance image; and use the contrast parameters and energy parameters as texture features.

[0269] In one optional implementation, the generation module 503 is specifically used to: determine the attribute category of the target object based on multi-dimensional features; if the attribute category of the target object belongs to the protection category, determine the spatial attitude of the target object based on the multi-dimensional features; and generate a control strategy for the target device based on the attribute category or spatial attitude of the target object.

[0270] In one optional implementation, the multi-dimensional features include: living features, which are signal features of the periodic physiological activities of a living organism; the generation module 503 is further configured to: determine whether the target object has life features based on the living features; if it has life features, determine the attribute category of the target object as a living organism; if it does not have life features, determine the attribute category of the target object as a non-living organism.

[0271] In one optional implementation, the generation module 503 is specifically configured to: if the attribute category of the target object is non-living, generate an inhibition command for the target device, the inhibition command being used to inhibit the protective action of the safety device in the target device; if the attribute category of the target object is child, generate a first control command for the target device, the first control command being used to trigger the target device to control with a first intensity parameter, the first intensity parameter being lower than the calibrated intensity parameter; if the attribute category of the target object is adult and the corresponding spatial posture is a standard posture, generate a second control command for the target device, the second control command being used to trigger the target device to control with the calibrated intensity parameter; if the attribute category of the target object is adult and the corresponding spatial posture is a forward-leaning posture, generate a third control command for the target device, the third control command being used to trigger the target device to control with the calibrated intensity parameter and adjust the control direction.

[0272] It should be noted that for details not disclosed in the control device of this application embodiment, please refer to the details disclosed in the control method of this application embodiment, which will not be repeated here.

[0273] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more microprocessors, or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SOC).

[0274] Optionally, embodiments of this application also provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor performs the steps of the control method for moving the storage medium described in the above embodiments. The specific implementation and technical effects are similar and will not be repeated here.

[0275] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the functional units in the various embodiments of this application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit described above can be implemented in hardware or in the form of hardware plus software functional units.

[0276] Optionally, this embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to implement a control method provided in the above embodiment.

[0277] In this embodiment, the device, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0278] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0279] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0280] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A control method, characterized in that, include: In response to a target object entering the sensing space of the projected capacitive sensor array of the target device, a corresponding capacitance image sequence is acquired, the capacitance image sequence including multiple images of the capacitance value of each coordinate point on the projected capacitive sensor array changing over time due to the target object; Multi-dimensional features are extracted from the capacitance image sequence, and these multi-dimensional features are features extracted from the temporal and spatial domains of the capacitance image sequence. Based on the aforementioned multi-dimensional features, a control strategy for the target device is generated.

2. The method according to claim 1, characterized in that, The step of parsing multi-dimensional features from the capacitance image sequence includes: The capacitance image sequence is processed to determine the projection area of ​​the target object on the projected capacitance sensor array; The multi-dimensional features are obtained by analyzing the capacitance image sequence corresponding to the projection region in both the temporal and spatial domains.

3. The method according to claim 2, characterized in that, The process of processing the capacitance image sequence to determine the projection area of ​​the target object on the projected capacitance sensor array includes: Based on a preset capacitance threshold, the capacitance value of each coordinate point in the capacitance image sequence is thresholded to obtain a binary image sequence. Each frame of the binary image sequence is analyzed to aggregate adjacent pixels with the same pixel value into a connected region. From each connected region, extract the connected region with the largest area as the projection region of the target object.

4. The method according to claim 2, characterized in that, The process of analyzing the capacitance image sequence corresponding to the projection region in both the temporal and spatial domains to obtain the multi-dimensional features includes: Based on the temporal changes of the capacitance image sequence corresponding to the projection region, the liveness features are obtained by analysis; Based on the spatial distribution of capacitance values ​​in the projection region in a single-frame capacitance image, spatial contour features are calculated. The multi-dimensional features include: the living features and the spatial contour features.

5. The method according to claim 4, characterized in that, The multidimensional features include: living features, which are signal features of the periodic physiological activities of a living organism; The method of analyzing the temporal changes of the capacitance image sequence corresponding to the projection region to obtain liveness features includes: Signal extraction is performed on the capacitance image sequence corresponding to the projection area to obtain the time-domain signal of the capacitance change in the projection area; Spectral analysis is performed on the time-domain signal to extract the periodic fluctuation characteristics of the time-domain signal within a preset vital sign frequency band; the periodic fluctuation characteristics are used as the living characteristics.

6. The method according to claim 4, characterized in that, The spatial contour features include: geometric features, which are the shape, size, and positional distribution of the projection area of ​​the target object on the projected capacitive sensor array; The calculation of spatial contour features based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image includes: Based on the single-frame capacitance image corresponding to the projection area, calculate the area of ​​the projection area and the coordinates of the center of gravity of the projection area on the projected capacitance sensor array. The area and the coordinates are used as the geometric features.

7. The method according to claim 4, characterized in that, The spatial contour features also include: statistical features, which are statistical parameters calculated based on the capacitance values ​​within the projection area of ​​the target object; The calculation of spatial contour features based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image includes: Based on the single-frame capacitance image corresponding to the projection area, calculate the mean and variance of the capacitance values ​​within the projection area; The mean and the variance are used as the statistical features.

8. The method according to claim 4, characterized in that, The spatial contour features also include: texture features, which are the features of the image texture pattern of the capacitance values ​​within the projection area of ​​the target object; The calculation of spatial contour features based on the spatial distribution of capacitance values ​​in the projection region within a single-frame capacitance image includes: Based on the single-frame capacitance image corresponding to the projection area, the corresponding gray-level co-occurrence matrix is ​​calculated; the gray-level co-occurrence matrix is ​​a joint probability distribution matrix of pixel gray values ​​that appear in pairs in a preset direction and at a preset interval, based on the gray levels of the capacitance image. Based on the gray-level co-occurrence matrix, contrast and energy parameters are extracted, where contrast represents the degree of local gray-level change in the capacitor image, and energy parameters represent the degree of gray-level distribution in the capacitor image. The contrast parameter and the energy parameter are used as the texture features.

9. The method according to claim 1, characterized in that, The process of generating a control strategy for the target device based on the multi-dimensional features includes: Based on the multi-dimensional features, the attribute category of the target object is determined; If the attribute category of the target object belongs to the protection category, then the spatial pose of the target object is determined based on the multi-dimensional features; A control strategy for the target device is generated based on the attribute category of the target object or the spatial attitude.

10. The method according to claim 9, characterized in that, The multidimensional features include: living features, which are signal features of the periodic physiological activities of a living organism; Determining the attribute category of the target object based on the multi-dimensional features includes: Based on the aforementioned living characteristics, it is determined whether the target object possesses life characteristics; If the target object exhibits signs of life, its attribute category is determined to be a living organism. If the target object does not possess any life characteristics, then its attribute category is determined to be non-living.

11. The method according to claim 9, characterized in that, The step of generating a control strategy for the target device based on the attribute category of the target object or the spatial attitude includes: If the attribute category of the target object is non-living, then an inhibition command for the target device is generated, which is used to inhibit the protective action of the safety device in the target device; If the attribute category of the target object is child, a first control command is generated for the target device. The first control command is used to trigger the target device to be controlled by a first intensity parameter, which is lower than the calibrated intensity parameter. If the attribute category of the target object is adult and the corresponding spatial posture is standard posture, then a second control command for the target device is generated. The second control command is used to trigger the target device to control according to the calibration intensity parameter. If the attribute category of the target object is adult and the corresponding spatial posture is forward tilt posture, then a third control command for the target device is generated. The third control command is used to trigger the target device to control with the calibration intensity parameter and adjust the control direction.

12. A vehicle, characterized in that, The vehicles include: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the vehicle to perform the method as described in any one of claims 1 to 11.

Images