A power control system for intelligent light-controlled products based on gesture recognition

Abstract

The present invention discloses a power control system for intelligent light-controlled products based on gesture recognition, which relates to the field of power supply technology for intelligent light-controlled products. The system includes an image enhancement module, a trajectory recognition module, a background modeling module, a path evaluation module, a control decision module, and an instruction execution module. The path evaluation module comprehensively evaluates the cleanliness of the user's gesture trajectory based on the numerical results of three types of data: the number of trajectory interruptions, the proportion of contaminated candidate segments, and the background light disturbance model. Based on the evaluation results, the current gesture is classified into three categories: valid gestures, suspicious gestures, and invalid gestures. The control decision module determines whether a power-on operation should be triggered based on the classification results of the user's current gestures, and performs the corresponding processing operations. The present invention solves the problem of gesture recognition misjudgment under high-brightness background interference and improves the accuracy and stability of power control response in complex lighting environments.

Description

Technical Field

[0001] The present invention relates to the technical field of power supply for intelligent light-controlled products, and in particular to a power supply control system for intelligent light-controlled products based on gesture recognition. Background Art

[0002] Smart light control products refer to lighting devices capable of automatic or remote lighting adjustment. A key element of these products is intelligent power control, enabling them to turn light sources on and off, and adjust brightness based on the environment, user behavior, or preset logic. As user demand for convenience, personalization, and contactless interaction continues to increase, traditional on-off control methods are increasingly unable to meet the evolving needs of smart lighting. Against this backdrop, gesture recognition-based control methods have emerged. This technology uses sensors to detect user hand movements, which are then analyzed by a processor to control power on / off or lighting parameters, delivering an intuitive, efficient, and contactless user experience. This not only enhances the user-friendly interface of lighting systems but also provides reliable technical support for intelligent upgrades in various scenarios, including homes and offices.

[0003] Existing gesture-recognition-based power control technologies for smart light-control products typically rely on real-time acquisition of user gestures using sensing devices such as infrared sensors, cameras, or millimeter-wave radar. After pre-processing, the collected gesture data is fed into an embedded processor or edge AI computing unit, where it extracts features and classifies them using pre-set image recognition algorithms or machine learning models. When the system recognizes a specific gesture command (such as waving, swiping, or pressing), it triggers a control logic module based on pre-defined gesture mappings. This control further controls actuators such as relays, thyristors, or electronic switches to turn the light-control product on and off or adjust its brightness. Furthermore, such systems typically include a power driver module, output control circuitry, and necessary filtering and protection circuitry to ensure stable command response and safe power usage. In some advanced systems, algorithms such as time window detection, gesture tracking, and background noise reduction are employed to improve recognition accuracy and false touch tolerance, resulting in a smoother and more accurate contactless control experience.

[0004] The existing technology has the following deficiencies:

[0005] When a user performs a light-on gesture in a bright environment, if there is a strobing or moving high-brightness light source in the background (such as car lights sweeping across curtains, mirror reflections, or mobile phone flashes), the gesture action path and the background dynamic light source will visually overlap. Since gesture recognition usually relies on the extraction of hand contours, movement trajectories, or grayscale change areas to determine whether the user has issued a control intention, dynamic light sources will form strong instantaneous bright spots or contour disturbances in the picture, thereby interfering with the extraction of hand features, causing the gesture image obtained by the recognition process on the key frame to be contaminated, resulting in the loss of valid features or the interruption of the trajectory. At this time, even if the user performs a complete light-on gesture, it may not be possible to form a control signal that the system can recognize, or it may mistakenly recognize the background bright spots as gesture trajectories and erroneously trigger the light-on operation. However, the existing power control technology for smart light control products based on gesture recognition cannot determine whether the power-on operation should be triggered based on the cleanliness of the user's gesture path in the presence of high-brightness background interference. As a result, the system can neither accurately respond to real operations nor shield false commands caused by background dynamic light sources, resulting in control failure and frequent false triggering, which seriously affects the stability and reliability of smart light control products in complex lighting environments.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0007] The purpose of the present invention is to provide a power control system for intelligent light-controlled products based on gesture recognition to solve the problems in the above-mentioned background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a power control system for an intelligent light-controlled product based on gesture recognition, comprising an image enhancement module, a trajectory recognition module, a background modeling module, a path evaluation module, a control decision module, and an instruction execution module;

[0009] The image enhancement module acquires a sequence of image frames during the user's gesture operation and performs local adaptive contrast enhancement processing on each frame of the image to enhance the edge contrast of the hand area and reduce the brightness of the highlight area in the image;

[0010] The trajectory recognition module extracts the gesture trajectory contour from the enhanced image frame sequence through edge detection and skin color recognition, counts the number of trajectory interruptions, and marks the trajectory segments with bright background interference in the extracted gesture trajectory contour as contamination candidate segments;

[0011] The background modeling module extracts background brightness change data from the non-gesture area of ​​the enhanced image frame and constructs a frame sequence-level background light perturbation model. The background light perturbation model includes three types of data: the average grayscale change amplitude, grayscale change slope, and highlight density of the background area.

[0012] The path evaluation module comprehensively evaluates the cleanliness of the user's gesture trajectory based on the numerical results of three types of data: the number of trajectory interruptions, the proportion of contaminated candidate segments, and the background light disturbance model. Based on the evaluation results, the current gesture is classified into three categories: valid gesture, suspicious gesture, and invalid gesture.

[0013] The control decision module determines whether to trigger the power-on operation based on the classification results of the user's current gesture and performs the corresponding processing operation;

[0014] The instruction execution module, after completing the power control operation, records the control execution results and gesture recognition process data, and updates the recognition behavior log for subsequent recognition model optimization.

[0015] Preferably, in the trajectory recognition module, in the enhanced image frame sequence, the image edge structure is extracted by edge detection, and the hand area is identified in the continuous frames based on skin color feature modeling, and the gesture trajectory profile is established according to the time sequence of the image frames; in the process of constructing the gesture trajectory profile, it is judged whether the trajectory jumps based on the position offset of the hand area in the adjacent frames. If a jump occurs, it is regarded as a trajectory interruption, and the number of trajectory interruptions is accumulated and counted; in the constructed gesture trajectory profile, the pixel brightness value of the trajectory segment overlapping with the highlighted area in the image is compared. If the average grayscale value of the trajectory segment area deviates from the normal hand grayscale range and the number of edge pixels is lower than the preset threshold, the trajectory segment is marked as a contaminated candidate segment.

[0016] Preferably, in the background modeling module, after excluding the hand area from the enhanced image frame sequence, background brightness change data of the non-gesture area in each frame of the image is extracted, and a brightness change data sequence of the non-gesture area is constructed according to the image frame sequence, where the background brightness change data is specifically pixel brightness values;

[0017] Calculate the average grayscale value change amplitude of the non-gesture area in multiple adjacent image frames according to the brightness change data sequence as the average grayscale value change amplitude of the background area;

[0018] Perform linear fitting on the average grayscale value change trend of the non-gesture area and calculate the change slope value within the unit frame time as the grayscale change slope of the background area;

[0019] In each frame, count the number of pixels in the non-gesture area whose brightness value is higher than the set threshold, and calculate the density value per unit area as the highlight density of the background area;

[0020] The average grayscale change amplitude, grayscale change slope and highlight density of the background area are used to form a background light disturbance model at the frame sequence level.

[0021] Preferably, in the path evaluation module, the method for obtaining the proportion of contaminated candidate segments is specifically as follows: counting the total number of image frames corresponding to the contaminated candidate segments, and counting the total number of image frames contained in the complete gesture trajectory contour; dividing the number of image frames corresponding to the contaminated candidate segments by the total number of image frames of the complete gesture trajectory contour to obtain the proportion of contaminated candidate segments.

[0022] Preferably, in the path evaluation module, the number of trajectory interruptions, the proportion of contaminated candidate segments, and the average grayscale change amplitude, grayscale change slope, and highlight density of the background area included in the background light disturbance model are normalized respectively; based on the results of the normalization processing, the gesture structure stability coefficient and the ambient light dynamic interference coefficient are generated; based on the generated gesture structure stability coefficient and the ambient light dynamic interference coefficient, the two are weightedly fused to generate a trajectory cleanliness index; a pre-set trajectory cleanliness index threshold range is determined, and after determination, it is compared with the generated trajectory cleanliness index, and the cleanliness of the user gesture trajectory is evaluated according to the comparison results, and the current gesture is divided into three categories: valid gesture, suspicious gesture, and invalid gesture according to the evaluation results.

[0023] Preferably, the acquisition logic of the gesture structure stability coefficient and the ambient light dynamic interference coefficient is as follows: the number of trajectory interruptions and the proportion of contaminated candidate segments are normalized respectively, specifically: the number of trajectory interruptions and the proportion of contaminated candidate segments are divided by their respective corresponding preset maximum values, and the normalized number of trajectory interruptions and the proportion of contaminated candidate segments are obtained respectively, and are calibrated as and , represents the normalized number of trajectory interruptions, represents the normalized proportion of contaminated candidate segments;

[0024] Calculate the gesture structure stability coefficient. The specific calculation formula is as follows: Where, is the gesture structure stability coefficient, and are all pre-set non-zero weight coefficients, and ;

[0025] The average grayscale change amplitude, grayscale change slope and highlight density of the background area included in the background light disturbance model are normalized respectively. Specifically, the three types of data, namely, the average grayscale change amplitude, grayscale change slope and highlight density of the background area included in the background light disturbance model, are divided by their respective corresponding preset maximum values ​​to obtain the normalized average grayscale change amplitude, grayscale change slope and highlight density of the background area, and are calibrated as 、 and , Indicates the average grayscale change of the normalized background area, Represents the grayscale change slope of the normalized background area, Represents the normalized highlight density of the background area;

[0026] Calculate the ambient light dynamic interference coefficient. The specific calculation formula is as follows: Where, is the ambient light dynamic interference coefficient, and are all pre-set non-zero weight coefficients, and .

[0027] Preferably, based on the generated gesture structure stability coefficient and ambient light dynamic interference coefficient, the two are weightedly fused to generate a trajectory cleanliness index. The specific calculation formula is as follows: Where, is the trajectory clean index, and are all preset non-zero weight coefficients, and .

[0028] Preferably, a preset trajectory clean index threshold interval is determined , and after determining the clean index of the generated trajectory The user's gesture trajectory is compared and evaluated based on the comparison results. The current gesture is divided into three categories: valid gesture, suspicious gesture, and invalid gesture based on the evaluation results. The specific comparison analysis and classification are as follows:

[0029] like , if the cleanliness level of the user’s gesture trajectory is low, the user’s current gesture is classified as an invalid gesture;

[0030] like ,If the cleanliness of the user’s gesture trajectory is medium, the user’s current gesture is classified as a suspicious gesture;

[0031] like , if the cleanliness of the user's gesture trajectory is high, the user's current gesture is classified as a valid gesture.

[0032] Preferably, in the control decision module, based on the classification result of the user's current gesture, it is determined whether the power-on operation should be triggered and the corresponding processing operation is performed, specifically:

[0033] When the user's current gesture is classified as a valid gesture, it is determined that a power-on operation should be triggered, a power-on control instruction is directly generated, and the control operation is executed;

[0034] When the user's current gesture is classified as a suspicious gesture, the system determines whether a power-on operation should be triggered based on the execution result statistics of the corresponding category of gesture trajectories recorded in the recognition behavior log within a preset period of time. If the number of successful executions exceeds a preset judgment threshold, a power-on control instruction is generated and the control operation is executed; otherwise, the power control operation is not executed.

[0035] When the user's current gesture is classified as an invalid gesture, it is determined that the power-on operation should not be triggered, and the power control process is terminated.

[0036] The technical effects of the present invention include:

[0037] This invention significantly improves the robustness and accuracy of gesture recognition systems in environments with strong background light interference. By introducing an image enhancement module to perform local adaptive contrast processing on image frames, the clarity of hand edge features is improved while the interference effect of highlight areas is weakened, laying a stable foundation for subsequent hand trajectory recognition. The trajectory recognition module uses a combination of edge detection and skin color modeling to accurately extract gesture trajectories in continuous image frames, and effectively identifies abnormal disturbance areas in the trajectory through trajectory interruption detection and highlight contamination segment calibration, improving the ability to restore hand trajectories under dynamic background interference, thereby solving the problems of high misrecognition rate and frequent trajectory interruptions in traditional solutions under complex lighting scenarios such as strong reflections and strobing lights.

[0038] The present invention has a highly quantifiable gesture trajectory "cleanliness" evaluation mechanism, which improves the intelligence and judgment accuracy of the system control decision-making. The path evaluation module introduces two normalized evaluation parameters, "gesture structure stability coefficient" and "ambient light dynamic interference coefficient", in the trajectory structure characteristics and environmental interference dimensions, and generates a trajectory cleanliness index through weighted fusion. On this basis, gestures are divided into three categories: valid, suspicious and invalid, achieving a leap forward from traditional image threshold-based judgment to multi-factor fusion modeling evaluation. At the same time, the system assists in judging the execution results of suspicious gestures through behavioral log records and historical statistical values ​​within a time period, thereby further avoiding false triggering and control delays, making the triggering of control instructions more consistent with the actual scene semantics, and ensuring the accuracy and stability of the system response;

[0039] The present invention establishes a closed-loop mechanism for the entire process from recognition to decision-making to feedback optimization, which helps to improve the sustainable adaptive capabilities of the entire system. The instruction execution module not only completes the generation of control signals and operation execution, but also records the execution results and gesture recognition process data in the behavior log, providing high-quality data support for the iterative optimization of subsequent recognition models. This mechanism ensures that the system continuously accumulates dynamic behavior data of recognition and control during multiple recognition processes, which can be used for model retraining, strategy adjustment or parameter fine-tuning, so as to achieve continuous learning and accurate response of the system in complex environments. At the same time, the software implementation path of this solution is clear, does not rely on specific hardware, has good engineering feasibility and cross-platform adaptability, and is suitable for promotion in multiple scenarios such as smart lighting and smart home. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a module schematic diagram of a power control system for an intelligent light-controlled product based on gesture recognition according to the present invention. DETAILED DESCRIPTION

[0041] The present invention provides Figure 1 The power control system of an intelligent light-controlled product based on gesture recognition shown in the figure includes an image enhancement module, a trajectory recognition module, a background modeling module, a path evaluation module, a control decision module, and an instruction execution module;

[0042] The image enhancement module acquires a sequence of image frames during the user's gesture operation and performs local adaptive contrast enhancement processing on each frame of the image to enhance the edge contrast of the hand area and reduce the brightness of the highlight area in the image;

[0043] The image enhancement module establishes a data interface with the device's front-end image acquisition unit (such as an infrared camera, RGB camera, or depth vision sensor). It calls the image acquisition API and configures a specific sampling period (e.g., 30 frames per second) and trigger window (e.g., motion detection triggers capture). It continuously captures consecutive image frames, indexed by timestamps, as the user performs gestures. The software dynamically caches the frame sequence using a buffer sliding window mechanism, automatically removing static or background frames and retaining only the consecutive frames strongly associated with the gesture. This process is performed asynchronously by a background thread, ensuring smooth and real-time acquisition. The final output is a chronologically ordered, spatially coherent image frame sequence for subsequent enhancement processing.

[0044] Local adaptive contrast enhancement is performed on each image frame, which can be achieved by combining histogram equalization with edge-preserving filtering. First, local histogram statistics are performed on the image through block-based region division (e.g., a local grid based on a sliding window), and the brightness distribution and contrast index of each sub-block area are calculated. Then, limited adaptive histogram equalization (such as the CLAHE algorithm) is applied to areas with concentrated brightness or highlights to improve the contrast of low and medium grayscales and suppress brightness overexposure. To avoid enhancing the edges of bright backgrounds, edge detection (such as Canny or Sobel) is combined to identify the preferred area of ​​the hand outline. Only this area is enhanced, while non-hand areas are subjected to brightness compression. This process can be completed in real time on an embedded platform or edge computing unit using an image processing framework (such as OpenCV), without relying on hardware filtering or illumination compensation.

[0045] Because when users perform gestures such as turning on the lights, their hand movements often occur in environments disturbed by natural light, reflective surfaces, or sudden high-brightness backgrounds, the hand edges in the original image can easily become blurred or broken due to insufficient brightness contrast or overlapping highlights, seriously affecting the accuracy of subsequent contour extraction and trajectory recognition. Existing technologies often use global enhancement strategies, which tend to simultaneously amplify background light interference and reduce the significance of hand features. Therefore, by adaptively enhancing hand edges and suppressing high-brightness backgrounds in local areas, the clarity of gesture structures in images can be effectively improved, and the probability of misrecognition caused by light disturbances during the feature extraction stage can be reduced. This image preprocessing strategy provides a stable and clean input data foundation for the entire gesture path judgment link, and is one of the key software support points to ensure the stable operation of the system in complex lighting scenarios.

[0046] The trajectory recognition module extracts the gesture trajectory contour from the enhanced image frame sequence through edge detection and skin color recognition, counts the number of trajectory interruptions, and marks the trajectory segments with bright background interference in the extracted gesture trajectory contour as contamination candidate segments;

[0047] In this embodiment, in the trajectory recognition module, in the enhanced image frame sequence, the image edge structure is extracted through edge detection, and the hand area is identified in the continuous frames based on skin color feature modeling, and the gesture trajectory profile is established according to the time sequence of the image frames. During the gesture trajectory profile construction process, it is determined whether the trajectory has jumped based on the position offset of the hand area in adjacent frames. If a jump occurs, it is regarded as a trajectory interruption, and the number of trajectory interruptions is accumulated and counted. In the constructed gesture trajectory profile, the pixel brightness values ​​of the trajectory segments overlapping with the highlighted areas in the image are compared. If the average grayscale value of the trajectory segment area deviates from the normal hand grayscale range and the number of edge pixels is lower than a preset threshold, the trajectory segment is marked as a contaminated candidate segment for subsequent trajectory cleanliness assessment.

[0048] To accurately extract the gesture trajectory contour from the enhanced image frame sequence, the trajectory recognition module uses software to perform a combined process of edge detection and skin color region identification. First, an edge detection method (such as the Canny or Sobel operator) is used to perform a gradient calculation on each image frame to extract significant contour lines and structural edges, particularly geometric features such as the fingertips and palm edges of the hand region. A skin color feature model is then constructed. For example, a threshold range is set in the YCbCr or HSV color space that corresponds to the Asian skin color distribution. This is then used to perform pixel-by-pixel analysis to identify skin color regions that may belong to the hand. Spatial overlap analysis of skin color regions and edge contours allows for a relatively accurate hand contour extraction within each frame. Next, based on the temporal order of the image frames, the extracted hand center position or geometric centroid is tracked within each frame. These points are then continuously connected in the temporal dimension using a coordinate sequence to construct a complete gesture trajectory contour. This approach can enhance trajectory coherence and noise immunity through algorithms such as Kalman filtering, inter-frame motion vector matching, or pixel tracking based on optical flow. This method is adopted because in real-world environments, user hand movements are subject to interference factors such as natural motion blur, occlusion, or lighting changes, making it difficult to stably identify gesture paths by relying solely on a single feature. Edge structures can provide shape constraints, and skin color models can improve the stability of regional positioning. The combination of the two can effectively improve the accuracy of trajectory construction, providing a stable and reliable input basis for subsequent trajectory interruption detection and contaminated segment determination.

[0049] During gesture trajectory contour construction, the system analyzes hand region position offsets based on the image frame sequence to determine whether the trajectory has experienced a jump and count the number of interruptions. Specifically, when constructing the gesture trajectory, the system records the position of the identified hand region center point (such as the geometric center of gravity or the center of the bounding rectangle) in each consecutive image frame and calculates the Euclidean or Manhattan distance between the center points in adjacent frames. If this distance exceeds a preset threshold (which can be set based on image resolution and expected gesture movement speed), a jump is determined to have occurred between the current and previous frames, indicating a break in the temporal continuity of the hand trajectory. Each detected jump is considered a trajectory interruption event, and the system simultaneously records the index of the interrupted frame in the trajectory data structure for subsequent cleanliness assessment. This statistical process is accomplished through a combination of simple coordinate difference calculations, sliding window filtering, and a dynamic threshold adjustment strategy to accommodate the varying gesture speeds of different users. Interruption detection and counting are necessary because in complex lighting conditions or unstable recognition environments, the hand image area may be obscured by a bright background or temporarily fail to be recognized, resulting in unnatural breaks in the trajectory. These breaks typically indicate that the gesture path is not being captured continuously, a significant sign of gesture image quality degradation. Therefore, by counting the number of trajectory interruptions, we can quantify the degree of discontinuity in the recognition chain, providing a key parameter for determining the cleanliness of the current gesture trajectory and effectively avoiding false triggering caused by misidentification interruptions.

[0050] To identify and demarcate trajectory segments affected by background noise within the constructed gesture trajectory profile, a software-based joint judgment process combining image brightness analysis and contour structure evaluation is performed. First, based on the previously extracted gesture trajectory, the system divides the entire trajectory into multiple continuous sub-segments, each corresponding to several image frames within a time window. Then, for each trajectory segment, the grayscale values ​​of all pixels within its spatial region are extracted, and the average grayscale value of the segment is calculated. To determine whether this brightness is abnormal, a "normal hand grayscale range" is pre-established as a reference. This range is obtained by statistically analyzing hand image samples of different skin tones under standard lighting conditions. It typically falls within the mid-grayscale range (e.g., 80-160 in 8-bit images) and can be dynamically adjusted based on the device's light sensitivity response curve. If the average grayscale value of the current trajectory segment deviates significantly from this normal grayscale range, it may be affected by background noise. The system also assesses the integrity of the segment's edge structure. This involves counting the number of edge pixels identified by the edge detection algorithm within the segment. If the number of edge pixels falls below a threshold set based on image resolution and typical hand size, the segment's structural features have been corrupted by light interference, preventing a clear hand outline from being formed. When these two conditions (abnormal grayscale values ​​and weakened edge structure) are met, the software logic designates the trajectory segment as a "contaminated candidate segment" and marks it in the trajectory data structure for subsequent use by the path evaluation module. This is done because bright backgrounds often introduce sudden bright spots or reflective areas in images, which can easily obscure the hand's grayscale information and hinder edge extraction. If these contaminated segments are not promptly identified and isolated, the system may mistakenly include non-hand areas in gesture recognition, resulting in misidentification or false triggering. This approach effectively quantifies and eliminates potentially contaminated segments, ensuring the accuracy and robustness of the trajectory evaluation phase.

[0051] The background modeling module extracts background brightness change data from the non-gesture area in the enhanced image frame and constructs a frame sequence-level background light perturbation model. The background light perturbation model includes three types of data: the average grayscale change amplitude, grayscale change slope, and highlight density of the background area. The background area refers to the non-gesture area formed after the gesture area is eliminated in the image processing stage, which is used to reflect the background lighting changes in the user's operating environment.

[0052] In this embodiment, in the background modeling module, after excluding the hand area from the enhanced image frame sequence, background brightness change data of the non-gesture area in each frame is extracted, and a brightness change data sequence of the non-gesture area is constructed according to the image frame sequence. The background brightness change data is specifically pixel brightness values.

[0053] In the background modeling module, to extract background brightness change data from non-gesture areas within the enhanced image frame sequence, software must first accurately exclude the hand region. This process relies on the gesture region mask information previously obtained by the trajectory recognition module. This mask is used to remove the corresponding hand pixels from each frame, retaining only the remaining non-gesture area for analysis. Subsequently, a grayscale extraction operation is performed on the pixels in the retained area within each frame, uniformly converting the image data from RGB or other color spaces to a grayscale image. The grayscale value (i.e., brightness value) of each pixel in the non-gesture area is recorded. By traversing the entire image frame sequence and collecting these brightness value data in chronological order, a "brightness change data sequence" for the non-gesture area is constructed, which reflects the temporal trend of background brightness. This process can be combined with a sliding window mechanism to further refine the granularity of inter-frame changes and improve temporal continuity. This approach is adopted because non-gesture areas often contain the user's background scene, and changes in its brightness directly reflect the dynamic behavior of ambient lighting disturbances. This is particularly true in scenes with strobe lights or moving bright sources (such as car lights and mirror reflections). These disturbances can manifest as noticeable fluctuations or sudden changes in the background brightness sequence. By excluding gesture areas and analyzing background brightness changes separately, we not only avoid interference caused by hand movements but also provide a clean, continuous, and quantifiable raw data foundation for the subsequent construction of background light disturbance models. This is a critical prerequisite for improving the system's ability to identify and judge dynamic light disturbances.

[0054] Calculate the average grayscale value change amplitude of the non-gesture area in multiple adjacent image frames according to the brightness change data sequence as the average grayscale value change amplitude of the background area;

[0055] To calculate the average grayscale variation in the non-gesture area across multiple adjacent image frames, software-based inter-frame difference analysis is performed based on a previously constructed brightness variation data sequence. Specifically, the system first averages the pixel grayscale values ​​in the non-gesture area within each frame to obtain the regional average grayscale value for the current frame. Then, within a sliding time window (e.g., containing 3, 5, or 7 frames), the system calculates the absolute difference between the regional average grayscale values ​​of two adjacent frames within the window, following the chronological order of the image frames. This difference represents the degree of brightness fluctuation between each frame and the previous one. The brightness variation between these adjacent frames is then accumulated and averaged to obtain the average grayscale variation in the non-gesture area within the time window. This metric reflects the overall fluctuation of background brightness in a short-term series and is an important means of quantifying ambient light stability. This approach is adopted because dynamic background light sources (such as flickering lights and moving reflections) often cause frequent brightness variations in the background area of ​​an image, and these variations directly affect image recognition reliability. This grayscale variation index can be used to determine whether the current background has obvious light disturbance characteristics, providing an accurate, real-time reference for subsequent path cleanliness assessment and false trigger avoidance. This processing can be performed simultaneously after image enhancement and trajectory recognition, and can be efficiently implemented using linear algorithms on edge computing devices or embedded platforms.

[0056] Perform linear fitting on the average grayscale value change trend of the non-gesture area and calculate the change slope value within the unit frame time as the grayscale change slope of the background area;

[0057] To calculate the slope of grayscale changes in non-gesture areas, software performs a linear fitting analysis of the average grayscale value trends in these areas. Specifically, the system first calculates the average grayscale value of each frame based on the pixel grayscale values ​​in the non-gesture area. These values ​​are then arranged chronologically to form a two-dimensional sequence of data points: frame time index - average grayscale value. A linear regression method (such as least squares fitting) is then used to fit this grayscale change sequence. The slope of the fitted line represents the direction and rate of change in average brightness per frame time, i.e., the grayscale change slope. A slope close to zero indicates relatively stable background brightness; a positive and large slope indicates increasing background brightness; a negative and rapidly changing slope indicates decreasing background light or occlusion. Linear fitting is used rather than simply observing instantaneous differences because real background light disturbances often have a certain degree of continuity and directionality. In scenarios such as stroboscopic light or when a light source slowly enters the field of view, a single-frame difference may not be sufficient to reflect the trend. Linear fitting, however, effectively smooths local fluctuations and extracts the direction of global changes, providing key support for subsequent determination of whether the trajectory is subject to continuous light interference. This processing, implemented on a sequence of image frames through simple linear algebra operations without relying on complex models, is suitable for deployment on edge devices and low-power platforms, offering excellent real-time and controllable performance.

[0058] In each frame, count the number of pixels in the non-gesture area whose brightness value is higher than the set threshold, and calculate the density value per unit area as the highlight density of the background area;

[0059] To calculate the density of highlight points in the background area, software performs threshold segmentation and statistical analysis of the pixel brightness in the non-gesture area of ​​each frame. This implementation first uses the previously identified gesture mask to exclude the hand area from each enhanced frame, retaining only the non-gesture area for brightness analysis. The system then extracts the brightness value of each pixel in this area and sets a highlight threshold (for example, 220 or higher for 8-bit images; the specific value can be determined based on the optical sensor's brightness response curve and the actual interference environment). All pixels with brightness values ​​above this threshold are considered highlight pixels. The system then counts these pixels and divides this number by the total number of pixels in the non-gesture area or the actual area of ​​the area to calculate the highlight point density per unit area. This density quantifies the concentration and distribution of strong light interference within the image and is a key indicator for determining whether strong background light sources (such as strong reflections or mirror flare) are intruding on the image. The significance of performing this calculation lies in the fact that in complex lighting environments, the presence of bright spots often indicates potential light spots or reflective disturbances. These areas can easily obscure gesture contours or forge contour boundaries, leading to recognition errors. By continuously monitoring the density of bright spots, not only can the assessment phase identify whether a gesture is located in a contaminated area, but it can also assist in identifying large-scale brightness flicker, thereby improving the robustness and adaptability of the entire gesture recognition system under changing background conditions. This entire process can be implemented using a low-latency algorithm during the preprocessing stage after image frame acquisition, making it suitable for embedded deployments for real-time recognition tasks.

[0060] The average grayscale change amplitude, grayscale change slope and highlight density of the background area are used to form a background light disturbance model at the frame sequence level.

[0061] To construct a frame-sequence-level background light disturbance model, software integrates and models the previously calculated background area's average grayscale change amplitude, grayscale change slope, and highlight density data. Specifically, the system first extracts the current values ​​of these three metrics within each sliding time window (e.g., 5 or 10 consecutive frames) to form a set of ternary feature vectors representing the comprehensive state of background light variation within that time period. The feature vectors corresponding to each window are then arranged chronologically within the frame sequence to form a time series model, which represents the dynamic representation of background light disturbances throughout the entire gesture execution cycle. To enhance the model's discriminative capabilities, the system can further perform smoothing, trend fitting, or interval segmentation statistics on the sequence to extract auxiliary information such as overall disturbance trends and abnormal fluctuation intervals for subsequent trajectory evaluation. The key significance of this modeling process lies in the fact that brightness anomalies in a single frame may be sporadic and unrepresentative. Fusion of multiple key metrics for analysis at the frame sequence level can significantly improve the ability to identify complex background disturbances. For example, when the average grayscale change amplitude is consistently high, the grayscale change slope is greater than zero, and the highlight density frequently increases, it can be determined that there is a persistent, strong interfering light source in the current background. This modeling result can be used to assess whether the user's gesture recognition environment is stable. This method relies entirely on software logic implementation, boasting excellent computational simplicity and real-time processing capabilities. It is suitable for gesture recognition control modules deployed in low-power smart lighting terminals, providing the system with dynamic, quantitative environmental intervention criteria in strong interference scenarios.

[0062] The path evaluation module comprehensively evaluates the cleanliness of the user's gesture trajectory based on the numerical results of three types of data: the number of trajectory interruptions, the proportion of contaminated candidate segments, and the background light disturbance model. Based on the evaluation results, the current gesture is classified into three categories: valid gesture, suspicious gesture, and invalid gesture.

[0063] In this embodiment, in the path evaluation module, the method for obtaining the proportion of contaminated candidate segments is specifically as follows: counting the total number of image frames corresponding to the contaminated candidate segments, and counting the total number of image frames contained in the complete gesture trajectory contour; dividing the number of image frames corresponding to the contaminated candidate segments by the total number of image frames of the complete gesture trajectory contour to obtain the proportion of contaminated candidate segments.

[0064] In the path assessment module, to determine the percentage of candidate contaminated segments, the image frame sequence and trajectory profile data structure are synchronized in time using software. First, during the gesture trajectory recognition phase, the system labels each image frame as a candidate contaminated segment. A list of Boolean labels corresponding to the time sequence is then created within the frame sequence, where each label identifies whether the frame is contaminated. Next, the system scans the list of image frames corresponding to the entire trajectory profile and counts the number of frames marked as contaminated, i.e., the number of contaminated segments. The total number of image frames covered by the entire gesture is also counted, i.e., the complete frame length of the gesture trajectory profile. Finally, the system divides the number of contaminated frames by the number of complete frames to determine the percentage of candidate contaminated segments. This approach requires no additional hardware and relies solely on the data structure generated during the trajectory recognition phase. By combining image frame sequence indices with label arrays, it efficiently and accurately calculates the percentage of contaminated segments at the software level, ensuring a quantifiable basis for subsequent trajectory quality assessments.

[0065] In this embodiment, in the path evaluation module, the number of trajectory interruptions, the proportion of contaminated candidate segments, and the average grayscale change amplitude, grayscale change slope, and highlight density of the background area included in the background light disturbance model are normalized respectively; based on the results of the normalization processing, the gesture structure stability coefficient and the ambient light dynamic interference coefficient are generated; based on the generated gesture structure stability coefficient and the ambient light dynamic interference coefficient, the two are weightedly fused to generate a trajectory cleanliness index; a pre-set trajectory cleanliness index threshold range is determined, and after determination, it is compared with the generated trajectory cleanliness index, and the cleanliness of the user gesture trajectory is evaluated based on the comparison results, and the current gesture is divided into three categories: valid gesture, suspicious gesture, and invalid gesture according to the evaluation results.

[0066] In the path assessment module, normalization is performed on the number of trajectory interruptions, the proportion of contaminated candidate segments, and the average grayscale change amplitude, grayscale change slope, and highlight density included in the background light perturbation model. This is to eliminate incomparability caused by differences in units, dimensions, or numerical distribution ranges among the various data sources, ensuring a consistent numerical scale for indicators from different sources when constructing the comprehensive assessment model. Since the number of trajectory interruptions is a discrete count value, the proportion is a percentage, and grayscale change parameters are derived from the image pixel grayscale distribution, their original numerical ranges and the data dynamic range may differ significantly. Directly incorporating these into weighted fusion can lead to weight imbalance or a single indicator overwhelming the overall assessment result, thus losing objective evaluation effectiveness. Through normalization, the system maps various indicators to standard intervals such as [0, 1] or [-1, 1]. This ensures equal scaling of different data within the computational model, providing a data foundation for the subsequent generation of structural stability coefficients and interference coefficients, ensuring consistent, stable, and interpretable assessment results.

[0067] In the path assessment module, to achieve the "pre-defined trajectory cleanliness index threshold range," software can dynamically set it by combining historical recognition data with statistical modeling methods. First, during the training phase or initial operation, the system collects a large number of actual gesture interaction samples, including valid gestures, false touch gestures, and invalid gestures performed by users under various lighting interference conditions. The corresponding trajectory cleanliness index is calculated for each sample. Then, using this sample data, cluster analysis or probability density estimation methods are used to identify the natural distribution boundaries of the trajectory cleanliness index between the three gesture categories (valid, suspicious, and invalid). For example, a Gaussian mixture model can be used to fit the overall trajectory cleanliness index distribution curve, further extracting the intersection area between categories as the critical value range. Two key thresholds can then be set: one for the cutoff between valid and suspicious gestures, and the other for the cutoff between suspicious and invalid gestures. These threshold ranges can be fixed in the model or continuously optimized through real-time updates to the recognition log to adapt to the differences in user behavior and usage scenarios, ensuring that the trajectory cleanliness classification results have reasonable discriminatory power and stability in actual operation.

[0068] In this embodiment, the logic for obtaining the gesture structure stability coefficient and the ambient light dynamic interference coefficient is as follows:

[0069] The number of trajectory interruptions and the proportion of contaminated candidate segments are normalized respectively. Specifically, the number of trajectory interruptions and the proportion of contaminated candidate segments are divided by their respective corresponding preset maximum values ​​to obtain the normalized number of trajectory interruptions and the proportion of contaminated candidate segments, and are marked as and , represents the normalized number of trajectory interruptions, represents the normalized proportion of contaminated candidate segments;

[0070] In the path assessment module, normalization of the number of trajectory interruptions and the proportion of contaminated candidate segments is implemented in software using a standard normalization formula. This involves dividing the raw values ​​by their respective preset maximum values ​​to map them to a uniform range of [0, 1]. This process relies on two statistical outputs from the previous recognition stage: the count of trajectory interruptions and the proportion of contaminated candidate segments, calculated as the ratio of the number of contaminated frames to the total number of gesture frames. During normalization, the system converts the current detection results based on a predefined maximum reference value to generate unitless, uniformly scaled results. This operation aims to eliminate differences in numerical scale and distribution dynamic range among different features, thereby preventing the "amplification" or "masking" of the evaluation results by a particular feature type during subsequent weighted modeling or multi-metric fusion evaluation. This normalization ensures logical comparability between trajectory interruptions and contamination levels and provides a sound mathematical foundation for integration, making subsequent evaluation results more stable, reliable, and versatile.

[0071] The corresponding preset maximum values ​​are set based on extensive experimental data during the system design phase or initial operation, and serve as reference limits for the normalization process. The maximum number of trajectory interruptions is determined by collecting multiple sets of typical user gesture data under complex lighting, occlusion, or background interference conditions, counting the extreme upper limits of interruption counts, and then setting a conservative and reasonable upper limit (e.g., 8 to 10 are common) based on the recognition error tolerance of the actual scenario. The maximum percentage of contaminated candidate segments is determined by statistical analysis and boundary preservation based on the upper limit of the proportion of contaminated segments in samples acquired in high-dynamic light interference environments (e.g., 80%). The preset maximum values ​​for both should be neither too low to avoid weakening the normalization discrimination nor too high to avoid distorting the actual evaluation. Ultimately, these maximum values ​​can be stored in the system parameter configuration file or model initialization parameters, used during runtime for normalization conversion, and dynamically adjusted based on runtime logs or automatically calibrated by the model training strategy.

[0072] Calculate the gesture structure stability coefficient. The specific calculation formula is as follows: Where, is the gesture structure stability coefficient, and are all pre-set non-zero weight coefficients, and ;

[0073] The design idea of ​​this formula is to comprehensively consider the two factors of "track breakage degree" and "track contamination degree" to evaluate the overall stability of the gesture structure. It represents the normalized number of trajectory interruptions. Its square term is used to amplify the negative impact of samples with a large number of interruptions on the results, reflecting the high sensitivity of trajectory continuity to gesture validity. represents the normalized proportion of candidate contaminated segments. Its square root is used to mitigate the nonlinear effects of extreme contamination ratios and to reflect the more gradual impact of the relative extent of background interference on stability. The combination of these two values ​​is placed in a "1 - [...]" structure, indicating that the system applies a negative penalty to the structural stability of the gesture. Lower scores indicate less stable structures, while higher scores indicate more complete, clearly defined, and more stable tracks. This formula not only suppresses the evaluation weight of highly fragmented and contaminated tracks, but also maintains the ability to express stability differences in a hierarchical manner through mathematical transformations, thus providing a solid structural scoring foundation for the subsequent generation of a track cleanliness index.

[0074] number and is a preset non-zero weight coefficient used to control the influence of trajectory breakage factor and pollution ratio factor on the gesture structure stability coefficient. The sum of them is always 1, which satisfies This type of linear normalization constraint helps ensure that both factors contribute to the final score, but does not expand the score range. and The specific value of can be determined through empirical statistics or machine learning training stage: During the system training stage, a large number of valid and invalid gesture samples of users in typical interference environments can be collected to calculate the proportion of trajectory breakage and contamination respectively, and the minimum error fitting, linear regression or objective function based on accuracy optimization can be used to determine the value of and Adjust the parameters so that the correlation between GSSC and the stability level of the gestures marked by humans is the strongest. Generally, if the system pays more attention to the continuity of the gesture, it can be given Higher weight; if the system is in an environment with frequent bright background interference, The weights can be increased appropriately. The final weight configuration reflects the actual degree of harm of the two types of problems while also taking into account the adjustability and versatility of the model. It can be solidified during the deployment phase or automatically adjusted through an adaptive mechanism.

[0075] The Gesture Structure Stability Coefficient (GSSC) is positively correlated with the cleanliness of the user's gesture trajectory. Higher GSSC values ​​indicate a more structurally coherent gesture trajectory, less susceptible to background interference, and overall cleaner. Conversely, lower GSSC values ​​indicate severe structural anomalies (such as frequent breaks) or a high concentration of contaminated segments, resulting in poor trajectory continuity and recognition integrity. GSSC evaluates the continuity and purity of a gesture path by combining the number of breakages and the proportion of contaminated segments. Its calculation incorporates a squared amplification of the number of breakages and a square root reduction of the proportion of contamination to accurately balance different types of anomalies. High GSSC values ​​typically indicate a smooth, clear gesture performed without significant background interference, thus more reliably reflecting true intent. Low GSSC values, on the other hand, often indicate that the system's recognized trajectory is perceptually contaminated or defective, making it unsuitable for triggering control. Therefore, this coefficient plays a core role in the subsequent trajectory cleanliness index generation and gesture effectiveness classification, and is one of the key indicators for determining whether to respond to user commands.

[0076] The average grayscale change amplitude, grayscale change slope and highlight density of the background area included in the background light disturbance model are normalized respectively. Specifically, the three types of data, namely, the average grayscale change amplitude, grayscale change slope and highlight density of the background area included in the background light disturbance model, are divided by their respective corresponding preset maximum values ​​to obtain the normalized average grayscale change amplitude, grayscale change slope and highlight density of the background area, and are calibrated as 、 and , Indicates the average grayscale change of the normalized background area, Represents the grayscale change slope of the normalized background area, Represents the normalized highlight density of the background area;

[0077] To unify the numerical scales of different interference indicators, the three types of data included in the background light disturbance model—the average grayscale variation amplitude of the background area, the grayscale variation slope, and the highlight density—must be normalized separately. This normalization can be performed using the "maximum normalization" method, which divides the current value of each indicator by its preset maximum value to normalize the result to the interval [0, 1]. This eliminates weight bias caused by differences in the original dimensions or magnitudes of different indicators. In implementation, after extracting background disturbance data, the system uses a software module to calculate the ratio of each actual value in the current frame sequence to the predefined maximum value, and caches the calculated results for subsequent coefficient generation. This operation not only simplifies the evaluation model's operational logic but also improves the comparability of various disturbance features in the evaluation function. This allows the contributions of different indicators to the trajectory cleanliness index to be uniformly weighed under the same weighting system, preventing a single data type from dominating the overall evaluation process due to its excessively high value.

[0078] The preset maximum value for each type of data is usually obtained through statistical analysis of a large amount of sample data. Specifically, in the early stage of model training or deployment, the system can collect a large amount of gesture interaction image frame data in a variety of real or simulated lighting disturbance environments, and extract the average grayscale change amplitude, grayscale change slope and highlight density of the corresponding background area. Subsequently, statistical modeling is performed on each indicator through histogram analysis or percentile method (such as taking the 95% percentile) after extreme value elimination, and a more reasonable maximum upper limit is set to ensure that most abnormal situations are covered and that extreme outliers avoid distortion of the normalized results. In addition, these maximum value parameters can be solidified in the system parameter file, or adaptively updated through log accumulation during model operation, so as to achieve dynamic adaptation to the lighting complexity in different application scenarios and ensure the stability, generalization and evaluation reliability of the normalization process.

[0079] Calculate the ambient light dynamic interference coefficient. The specific calculation formula is as follows: Where, is the ambient light dynamic interference coefficient, and are all pre-set non-zero weight coefficients, and .

[0080] The core purpose of this formula is to comprehensively measure the interference risk that background light changes may cause to the gesture recognition process. is the normalized average grayscale variation, which is used to characterize the degree of fluctuation of light intensity; is the normalized grayscale change slope, reflecting the speed of light change. The product of the two represents the overall intensity of severe light disturbance, while the square root processing can suppress the amplification effect of extreme interference values ​​on the results while retaining the change trend, making the results more stable and representative. =(ln0) is the normalized highlight density, reflecting the spatial concentration of high-intensity bright spots in the scene. Adding one to it and then taking its logarithm avoids the risk of undefined logarithms (e.g., ln0) and smoothes the growth of the influence of high-density areas, preventing linear growth from causing nonlinear amplification of the final result. Overall, this composite calculation method captures the intensity, speed, and distribution characteristics of dynamic ambient light disturbances while maintaining excellent stability and sensitivity, providing a solid quantitative foundation for subsequent trajectory cleanliness assessment.

[0081] and They are the weight coefficients of the two components in the calculation of the ambient light dynamic interference coefficient, which are used to balance the contribution ratio of the two types of interference factors in the final evaluation. For control and The degree of influence of the joint term on ELDC, Control The sum of the two is 1, ensuring that the entire coefficient calculation logic is carried out within a unified weight system, which is convenient for subsequent model optimization and interpretation. The specific values ​​of these weight coefficients are usually derived from the statistical analysis of a large amount of labeled data and the model performance evaluation process during the training phase. For example, different machine learning optimization methods such as grid search and cross validation can be used to test different and The system combines performance indicators such as recognition accuracy and error rate in real or simulated environments, ultimately selecting the optimal combination as the preset coefficients. Furthermore, to enhance the system's adaptability to different application scenarios, these two weights can be set as adjustable parameters, dynamically updated during deployment or operation to account for changes in the main interference factors in different lighting environments, thereby maintaining system recognition accuracy and robustness.

[0082] The value of the ambient light dynamic interference coefficient (ELDC) is negatively correlated with the "evaluation of the cleanliness of the user's gesture trajectory", that is, the higher the ELDC value, the more drastic the lighting changes in the environment during the current gesture execution, the stronger the dynamic highlight interference in the background, and the greater the probability that the gesture image is contaminated by the background, resulting in a lower degree of cleanliness of the gesture trajectory. Specifically, if the product term in ELDC If the value is large, it means that the brightness of the background area changes not only in a large amplitude but also at a fast speed, and there may be frequent interference scenes such as light spots passing by and light flashing. Large values ​​indicate a dense distribution of pixels with excessive brightness in the non-gesture area, and the presence of significant spatial interference sources such as reflections and bright backgrounds. These factors interfere with the stable extraction of hand contours and trajectory features in the image, leading to trajectory interruptions and misrecognition. Therefore, as a quantitative indicator of the degree of dynamic disturbance in ambient light, higher ELDC values ​​indicate a greater likelihood of "contamination" in the gesture path, thus inhibiting the assessment of trajectory cleanliness and becoming a key parameter in determining whether a gesture is valid.

[0083] In this embodiment, based on the generated gesture structure stability coefficient and ambient light dynamic interference coefficient , the two are weightedly fused to generate the trajectory clean index. The specific calculation formula is as follows: Where, is the trajectory clean index, and are all preset non-zero weight coefficients, and .

[0084] In the path evaluation module, in order to calculate the trajectory cleanliness index, the system first obtains the values ​​of the gesture structure stability coefficient (GSSC) and the ambient light dynamic interference coefficient (ELDC), and then takes the complementary value of ELDC. , to enhance its directivity compared with GSSC. Then, the system linearly combines the two through weighted fusion to obtain the trajectory clean index (TCI), where the fusion coefficient and is a preset non-zero weight coefficient that reflects the relative importance the system places on gesture structure and environmental interference. In specific implementation, these two coefficients can be obtained through empirical training of historical sample data: after collecting a large number of labeled recognition cases (including valid, suspicious, and invalid gestures) in the early stage, the different and The classification accuracy of the recognition results under each parameter combination is determined by selecting the weighted combination with the highest overall evaluation accuracy as the system's default configuration. Adjustable parameters can also be designed to adapt the tolerance strategy for false touches or missed recognitions in different scenarios. This approach allows the system to balance robustness and environmental adaptability by integrating two different types of indicators (structural and optical interference), effectively determining the cleanliness of gesture trajectories.

[0085] In this embodiment, a preset trajectory clean index threshold interval is determined. , and after determining the clean index of the generated trajectory The user's gesture trajectory is compared and evaluated based on the comparison results. The current gesture is divided into three categories: valid gesture, suspicious gesture, and invalid gesture based on the evaluation results. The specific comparison analysis and classification are as follows:

[0086] like , if the cleanliness level of the user’s gesture trajectory is low, the user’s current gesture is classified as an invalid gesture;

[0087] This situation indicates that the overall trajectory of the user's gesture has been severely disturbed. This is manifested in frequent interruptions in the trajectory structure, a high proportion of fragments in the gesture path covered by strong background light, and drastic light fluctuations in the background area itself. This means that the gesture signal obtained by the system during the recognition process has been severely distorted and cannot accurately restore the user's true intention. If a power control operation is triggered in this situation, it is very likely to cause an erroneous response. Therefore, the gesture should be judged as an invalid gesture to avoid erroneous activation of the power device and ensure safety and control accuracy.

[0088] like ,If the cleanliness of the user’s gesture trajectory is medium, the user’s current gesture is classified as a suspicious gesture;

[0089] This situation indicates that gesture recognition is subject to some degree of interference, but not to the point of complete failure. This typically means that while the gesture trajectory is partially affected by background light disturbances, such as slight interruptions or localized highlight contamination in key areas, the overall structure still has some recognition value. In this case, the system may classify the gesture as "suspicious," requiring further analysis based on user behavior history or environmental context. This processing strategy can improve the system's fault tolerance and recognition robustness in complex lighting conditions while ensuring safety.

[0090] like , if the cleanliness of the user's gesture trajectory is high, the user's current gesture is classified as a valid gesture.

[0091] This indicates that the user's gesture trajectory has a high degree of clarity and continuity throughout the entire sequence of image frames, and the influence of background light disturbances is negligible. The extracted hand contour is complete, the trajectory is coherent, and there are very few interference fragments. The system can be highly confident that the recognized action signal is a true and valid user control intention. Therefore, this gesture can be classified as a "valid gesture" and directly trigger the power-on command, thereby achieving fast and accurate contactless control operations. This judgment mechanism is the core guarantee of system reliability and helps to provide a stable and efficient user interaction experience.

[0092] The control decision module determines whether to trigger the power-on operation based on the classification results of the user's current gesture and performs the corresponding processing operation;

[0093] In this embodiment, in the control decision module, based on the classification results of the user's current gesture, it is determined whether the power-on operation should be triggered and corresponding processing operations are performed, specifically:

[0094] When the user's current gesture is classified as a valid gesture, it is determined that a power-on operation should be triggered, a power-on control instruction is directly generated, and the control operation is executed;

[0095] When the user's current gesture is classified as a valid gesture, the system can generate and execute a power-on control command through software. The specific process includes the following: First, the system determines that the gesture belongs to the valid category based on the gesture classification result output by the previous path evaluation module, and triggers the power control judgment logic. Next, a conditional judgment statement is set in the power control judgment logic to respond positively to the "valid gesture" label, calling the control interface to generate a control command string in a standard format, such as "POWER_ON". This control command can be sent to the control hardware end of the smart light control product through a communication interface (such as GPIO, serial port or I²C bus), driving the relay, switch chip or thyristor component to complete the power-on action. After the control command is executed, the system will also record the command trigger time, execution status (success or failure), and key parameters of the gesture recognition process in the background for subsequent recognition behavior log maintenance and model optimization. This design is based on the fact that "valid gestures" have already been determined in prior evaluations to possess structural integrity, low interference, and high confidence. Further confirmation or judgment would result in user operation delays or a reduced interactive experience. Therefore, directly triggering control commands for these gestures helps improve system responsiveness and intelligence, meeting the dual requirements of real-time and accuracy for intelligent light control products. This solution achieves an efficient closed-loop from gesture recognition to control execution through clear logical branching and interface calls. It can be fully implemented in software processes and adapted to different control platforms.

[0096] When the user's current gesture is classified as a suspicious gesture, the system determines whether a power-on operation should be triggered based on the execution result statistics of the corresponding category of gesture trajectories recorded in the recognition behavior log within a preset period of time. If the number of successful executions exceeds a preset judgment threshold, a power-on control instruction is generated and the control operation is executed; otherwise, the power control operation is not executed.

[0097] When the user's current gesture is classified as a suspicious gesture, the system determines whether to trigger the power-on operation by retrieving historical data from the recognition behavior log. The software implementation is as follows: After the system determines it is a "suspicious gesture," it first indexes the historical trajectory records that match this category of gesture within a preset time period (such as the past 24 hours or the last 100 records) in the log module based on the current time. Each record contains the evaluation classification of the gesture trajectory and a marking field indicating whether the control instruction was successfully executed. The system counts the number of successful historical executions under this gesture classification and compares it with a preset judgment threshold (such as the number of successful executions must be ≥5 times). If the trigger condition is met, a standard control command (such as "POWER_ON") is generated by calling the control interface and sent to the power control component of the light control product to complete the conduction operation. Otherwise, the system automatically terminates the control process and records the decision result of "not triggered this time" for subsequent optimization.

[0098] The fundamental reason for adopting this approach is that "suspicious gestures" themselves can be somewhat disruptive or incomplete, posing the risk of false triggering of power control, yet the possibility of valid input cannot be completely ruled out. Therefore, by statistically analyzing historical performance, a data-driven approach is employed to enhance the contextual awareness of decision-making, avoiding misjudgments caused by occasional anomalies and implementing a dynamic judgment strategy that balances fault tolerance and accuracy. This approach establishes a decision-making backtracking mechanism based on recognition behavior logs, ensuring the control logic's adaptability and learning capabilities. The entire process can be implemented entirely through software data processing and conditional judgment.

[0099] When the user's current gesture is classified as an invalid gesture, it is determined that the power-on operation should not be triggered, and the power control process is terminated.

[0100] When the user's current gesture is classified as an invalid gesture, the system uses software control logic to determine that the power-on operation should not be triggered. This is achieved as follows: After the system completes the gesture cleanliness assessment and classifies the current gesture as an "invalid gesture," it immediately passes the classification result to the control judgment logic. This logic has a clear rule that once the "invalid" label is received, the current user operation is assumed to be inexecutable, and no further historical verification or control instruction generation is required. The system skips the call to the power control interface through a conditional judgment statement, marks this operation as "process termination," and records the recognition event in the recognition behavior log, including the timestamp, gesture image frame sequence number, trajectory evaluation result, and termination reason field.

[0101] This processing method was designed to ensure accurate system responses and robustness against interference. "Invalid gestures" typically indicate a missing user action, abnormal gesture input, or severe background interference. Continuing control can lead to the risk of misoperation. Therefore, directly terminating the control process not only prevents false triggering but also conserves processing resources and improves system efficiency. Furthermore, this method relies entirely on software logic and logging modules, facilitating subsequent analysis and model optimization while maintaining system scalability and maintainability.

[0102] The instruction execution module, after completing the power control operation, records the control execution results and gesture recognition process data, and updates the recognition behavior log for subsequent recognition model optimization.

[0103] After completing a power control operation, the system automatically records two core data types related to the current operation through software: the power control execution result (i.e., whether the command was successfully executed and whether there were any false touches, etc.); and the complete gesture recognition process data before the control operation, including the trajectory image frame sequence, the number of trajectory interruptions, the proportion of contaminated candidate segments, the background light perturbation model value, the trajectory cleanliness index, and the final gesture category determination. This data is collected and packaged in real time through the background data interface or cache module and synchronously stored in the recognition behavior log database. The log uses a structured record format, enabling subsequent retrospective analysis and model training based on fields such as gesture number, execution status, and recognition parameters.

[0104] The core purpose of performing the above-mentioned recording and updating operations is to establish a "feedback learning mechanism" that enables the system to continuously optimize the recognition model by accumulating recognition results under real-world operating behaviors and environmental interference. For example, if a certain type of suspicious gesture is stably and successfully executed in multiple recognitions, the system can dynamically adjust the trajectory cleanliness index threshold; or if a certain type of false touch scenario frequently appears in the recognition behavior log, it can also be fed back to the image enhancement or trajectory filtering module for parameter fine-tuning. This mechanism improves the system's adaptability to user usage habits and complex lighting backgrounds, gradually enhancing the intelligence and personalization of the control logic, thereby significantly improving the recognition accuracy and control reliability of smart light control products in non-ideal environments.

[0105] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0106] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0107] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0108] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A power control system for intelligent light-controlled products based on gesture recognition, characterized in that: It includes image enhancement module, trajectory recognition module, background modeling module, path evaluation module, control decision module and instruction execution module; The image enhancement module acquires a sequence of image frames during the user's gesture operation and performs local adaptive contrast enhancement processing on each frame of the image to enhance the edge contrast of the hand area and reduce the brightness of the highlight area in the image; The trajectory recognition module extracts the gesture trajectory contour from the enhanced image frame sequence through edge detection and skin color recognition, counts the number of trajectory interruptions, and marks the trajectory segments with bright background interference in the extracted gesture trajectory contour as contamination candidate segments; The background modeling module extracts background brightness change data from the non-gesture area of ​​the enhanced image frame and constructs a frame sequence-level background light perturbation model. The background light perturbation model includes three types of data: the average grayscale change amplitude, grayscale change slope, and highlight density of the background area. The path evaluation module comprehensively evaluates the cleanliness of the user's gesture trajectory based on the numerical results of three types of data: the number of trajectory interruptions, the proportion of contaminated candidate segments, and the background light disturbance model. Based on the evaluation results, the current gesture is classified into three categories: valid gesture, suspicious gesture, and invalid gesture. In the path assessment module, the number of trajectory interruptions, the proportion of contaminated candidate segments, and the average grayscale change amplitude, grayscale change slope, and highlight density of the background area included in the background light disturbance model are normalized. Based on the normalized results, the gesture structure stability coefficient and the ambient light dynamic interference coefficient are generated. Based on the generated gesture structure stability coefficient and ambient light dynamic interference coefficient, the two are weightedly fused to generate a trajectory cleanliness index. A pre-set trajectory cleanliness index threshold range is determined and compared with the generated trajectory cleanliness index after determination. The cleanliness of the user's gesture trajectory is evaluated based on the comparison results, and the current gesture is classified into three categories based on the evaluation results: valid gesture, suspicious gesture, and invalid gesture. The logic for obtaining the gesture structure stability coefficient and the ambient light dynamic interference coefficient is as follows: The number of trajectory interruptions and the proportion of contaminated candidate segments are normalized respectively. Specifically, the number of trajectory interruptions and the proportion of contaminated candidate segments are divided by their respective corresponding preset maximum values ​​to obtain the normalized number of trajectory interruptions and the proportion of contaminated candidate segments, and are marked as and , represents the normalized number of trajectory interruptions, represents the normalized proportion of contaminated candidate segments; Calculate the gesture structure stability coefficient. The specific calculation formula is as follows: Where, is the gesture structure stability coefficient, and are all pre-set non-zero weight coefficients, and ; The average grayscale change amplitude, grayscale change slope and highlight density of the background area included in the background light disturbance model are normalized respectively. Specifically, the three types of data, namely, the average grayscale change amplitude, grayscale change slope and highlight density of the background area included in the background light disturbance model, are divided by their respective corresponding preset maximum values ​​to obtain the normalized average grayscale change amplitude, grayscale change slope and highlight density of the background area, and are calibrated as 、 and , Indicates the average grayscale change of the normalized background area, Represents the grayscale change slope of the normalized background area, Represents the normalized highlight density of the background area; Calculate the ambient light dynamic interference coefficient. The specific calculation formula is as follows: Where, is the ambient light dynamic interference coefficient, and are all pre-set non-zero weight coefficients, and ; The control decision module determines whether to trigger the power-on operation based on the classification results of the user's current gesture and performs the corresponding processing operation; The instruction execution module, after completing the power control operation, records the control execution results and gesture recognition process data, and updates the recognition behavior log for subsequent recognition model optimization.

2. The power control system for intelligent light-controlled products based on gesture recognition according to claim 1, characterized in that: In the trajectory recognition module, in the enhanced image frame sequence, the image edge structure is extracted through edge detection, and the hand area is identified in the continuous frames based on skin color feature modeling, and the gesture trajectory profile is established according to the chronological order of the image frames. During the gesture trajectory profile construction process, the position offset of the hand area in adjacent frames is used to determine whether the trajectory has jumped. If a jump occurs, it is regarded as a trajectory interruption, and the number of trajectory interruptions is accumulated and counted. In the constructed gesture trajectory profile, the pixel brightness values ​​of the trajectory segments overlapping with the highlighted areas in the image are compared. If the average grayscale value of the trajectory segment area deviates from the normal hand grayscale range and the number of edge pixels is lower than the preset threshold, the trajectory segment is marked as a contaminated candidate segment.

3. The power control system for intelligent light-controlled products based on gesture recognition according to claim 2, characterized in that: In the background modeling module, after excluding the hand area from the enhanced image frame sequence, the background brightness change data of the non-gesture area in each frame is extracted, and a brightness change data sequence of the non-gesture area is constructed according to the image frame sequence. The background brightness change data is specifically the pixel brightness value; Calculate the average grayscale value change amplitude of the non-gesture area in multiple adjacent image frames according to the brightness change data sequence as the average grayscale value change amplitude of the background area; Perform linear fitting on the average grayscale value change trend of the non-gesture area and calculate the change slope value within the unit frame time as the grayscale change slope of the background area; In each frame, count the number of pixels in the non-gesture area whose brightness value is higher than the set threshold, and calculate the density value per unit area as the highlight density of the background area; The average grayscale change amplitude, grayscale change slope and highlight density of the background area are used to form a background light disturbance model at the frame sequence level.

4. The power control system for intelligent light-controlled products based on gesture recognition according to claim 3, characterized in that: In the path evaluation module, the proportion of contaminated candidate segments is obtained by counting the total number of image frames corresponding to the contaminated candidate segments and counting the total number of image frames contained in the complete gesture trajectory outline; dividing the number of image frames corresponding to the contaminated candidate segments by the total number of image frames of the complete gesture trajectory outline to obtain the proportion of contaminated candidate segments.

5. The power control system for intelligent light-controlled products based on gesture recognition according to claim 4, characterized in that: Stability coefficient of gesture structure based on generation and ambient light dynamic interference coefficient , the two are weightedly fused to generate the trajectory clean index. The specific calculation formula is as follows: Where, is the trajectory clean index, and are all preset non-zero weight coefficients, and .

6. The power control system for intelligent light-controlled products based on gesture recognition according to claim 5, characterized in that: Determine the pre-set trajectory clean index threshold range , and after determining the clean index of the generated trajectory The user's gesture trajectory is compared and evaluated based on the comparison results. The current gesture is divided into three categories: valid gesture, suspicious gesture, and invalid gesture based on the evaluation results. The specific comparison analysis and classification are as follows: like , if the cleanliness level of the user’s gesture trajectory is low, the user’s current gesture is classified as an invalid gesture; like ,If the cleanliness of the user’s gesture trajectory is medium, the user’s current gesture is classified as a suspicious gesture; like , if the cleanliness of the user's gesture trajectory is high, the user's current gesture is classified as a valid gesture.

7. The power control system for intelligent light-controlled products based on gesture recognition according to claim 6, characterized in that: In the control decision module, based on the classification results of the user's current gesture, it is determined whether the power-on operation should be triggered and the corresponding processing operations are performed, specifically: When the user's current gesture is classified as a valid gesture, it is determined that a power-on operation should be triggered, a power-on control instruction is directly generated, and the control operation is executed; When the user's current gesture is classified as a suspicious gesture, the system determines whether a power-on operation should be triggered based on the execution result statistics of the corresponding category of gesture trajectories recorded in the recognition behavior log within a preset period of time. If the number of successful executions exceeds a preset judgment threshold, a power-on control instruction is generated and the control operation is executed; otherwise, the power control operation is not executed. When the user's current gesture is classified as an invalid gesture, it is determined that the power-on operation should not be triggered, and the power control process is terminated.

Images