Processing positioning method and system for adaptive image recognition

Abstract

The embodiment of the invention relates to the technical field of image processing and positioning, in particular to a processing and positioning method and system for adaptive image recognition, and the method comprises the steps: obtaining a multi-source image data set of a target processing scene; performing image state adaptability analysis processing on the original image sequence to obtain an image recognition feature set of each image region in the original image sequence; performing positioning state mapping processing on the image recognition feature set based on the equipment positioning state data set, and generating a spatial positioning driving feature set corresponding to each image region; and generating a self-adaptive positioning control instruction according to the spatial positioning driving feature set, transmitting the self-adaptive positioning control instruction to an execution end of the processing equipment, and indicating the processing equipment to execute self-adaptive pose positioning control operation for the current image area. Therefore, the flexibility of processing equipment to cope with complex processing scenes is greatly improved, and it is ensured that target processing operation is accurately executed under various observation poses and environmental conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and positioning, and more particularly to a processing positioning method and system for adaptive image recognition. Background Art

[0002] With the continuous development of the manufacturing industry, the requirements for machining accuracy and automation are increasing. As a key support for achieving precision machining, image recognition technology has been widely used in various machining scenarios. By acquiring and analyzing images of the machining scene, it aims to determine the exact position and posture of the machining equipment, ensuring the correct execution of the machining task.

[0003] However, existing image recognition and processing positioning technologies have obvious flaws. For example, most methods obtain single image data, relying only on a fixed perspective or limited data types, and are unable to fully reflect the actual situation of the processing scene. For another example, in the image analysis and processing link, there is a lack of sufficient consideration of environmental interference and changes in image status, resulting in one-sided features extracted and unable to accurately reflect the complex relationship between the processing object and the environment. In addition, when converting image features into control instructions for processing equipment, the existing technology is simple and crude, making it difficult to achieve adaptive adjustment of processing equipment for changing scenes. The processing accuracy and efficiency are limited, making it difficult to meet the requirements of complex processing tasks. Summary of the Invention

[0004] In view of this, the present invention provides a processing positioning method and system for adaptive image recognition.

[0005] An embodiment of the present invention provides a processing positioning method for adaptive image recognition, which is applied to a processing positioning system, the method comprising: acquiring a multi-source image data set of a target processing scene, the multi-source image data set comprising a sequence of original images collected under different observation postures of a processing device and a corresponding device positioning state data set; performing image state adaptive analysis processing on the original image sequence to obtain an image recognition feature set for each image area in the original image sequence, the image recognition feature set comprising spatial distribution features, image analysis features and environmental interference features; performing positioning state mapping processing on the image recognition feature set based on the device positioning state data set to generate a spatial positioning drive feature set corresponding to each image area; the spatial positioning drive feature set is used to characterize the spatial trajectory adjustment amount and posture correction direction required for the processing device to perform the target processing operation under the corresponding observation posture; generating an adaptive positioning control instruction based on the spatial positioning drive feature set and transmitting it to the execution end of the processing device, instructing the processing device to perform an adaptive posture positioning control operation for the current image area.

[0006] The present invention also provides a processing positioning system, comprising: a memory for storing program instructions and data; a processor for coupling with the memory and executing instructions in the memory to implement the above method.

[0007] The present invention also provides a computer storage medium comprising instructions, which implement the above method when executed on a processor.

[0008] The embodiment of the present invention integrates multi-dimensional data information of the processing scene by acquiring a multi-source image data set covering the original image sequence and the equipment positioning state data set of the target processing scene, so that subsequent analysis is based on a comprehensive data set; then, the original image sequence is adaptively analyzed and processed to obtain an image recognition feature set including spatial distribution, image analysis and environmental interference features, which can deeply analyze the properties of the image region from different key dimensions to accurately characterize the unique characteristics of each image region; then, the image recognition feature set is subjected to positioning state mapping processing based on the equipment positioning state data set to generate a spatial positioning drive feature set. This process realizes the effective mapping from image features to the operation adjustments required for the processing equipment, providing a key basis for the precise control of the processing equipment; finally, an adaptive positioning control instruction is generated based on the set and transmitted to the execution end of the processing equipment, so that the processing equipment can adaptively adjust the posture for different image areas, greatly improving the flexibility of the processing equipment in dealing with complex processing scenarios, ensuring the precise execution of target processing operations under a variety of observation postures and environmental conditions, improving processing accuracy and quality, reducing the need for manual intervention, and optimizing the overall processing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0010] Figure 1 A schematic flow chart of the steps of a processing positioning method for adaptive image recognition provided by an embodiment of the present invention.

[0011] Figure 2 This is a structural block diagram of a processing positioning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The technical solutions of the present invention will be described below in conjunction with the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation methods described in the following exemplary embodiments do not represent all implementation methods consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention. It should be noted that the terms "first", "second", etc. in the specification of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0013] See also Figure 1 , Figure 1 1 is a flow chart of a processing positioning method based on adaptive image recognition provided by an embodiment of the present invention. The method is applied to a processing positioning system and may further include steps 110 to 140.

[0014] Step 110: Acquire a multi-source image data set of a target processing scene, wherein the multi-source image data set includes a sequence of original images acquired at different observation postures of the processing equipment and a corresponding equipment positioning state data set.

[0015] In this embodiment of the present invention, the processing of a robot vacuum housing is used as an example. A processing workshop contains multiple processing equipment, and the target processing scene is the processing area surrounding the robot vacuum housing. The processing equipment includes a robotic arm and various processing tools. The equipment can operate the robot vacuum housing at different positions and angles. During this process, multi-source image data sets are required to assist in precise processing.

[0016] One of the multi-source image datasets in this embodiment is a sequence of raw images captured from different viewing positions of the processing equipment, specifically images captured from various angles and positions during the processing of the robot vacuum shell. Another corresponding dataset, the device positioning state dataset, records information such as the position and posture of the processing equipment at the time. The raw image sequence provides a visual representation of the processing scene and the shell's status, while the device positioning state dataset provides the foundation for subsequently linking image features with device operations.

[0017] When collecting the original image sequence, multiple cameras installed in different positions start working and continuously collect images from various perspectives; at the same time, sensors and other components on the device record the device's positioning status data in real time. This data includes device coordinates, rotation angles, etc., and finally forms a complete multi-source image data set.

[0018] As an implementation method, an original image sequence in a multi-source image data set of a target processing scene is obtained, including: Step 111: Acquire a real-time working status data set of a processing device, and analyze the device motion trajectory data and current processing task parameters in the real-time working status data set.

[0019] Taking the processing of a robot vacuum shell as an example, the processing equipment is the mechanical device responsible for the polishing and assembly processes. During operation, the equipment continuously generates real-time working status data, including the equipment's motion trajectory data, such as the robot arm's movement path in three-dimensional space, stored as a sequence of coordinate points. The current processing task parameters specify the specific requirements for the robot vacuum shell processing, such as the polishing accuracy standard and assembly position requirements. This data is recorded and generated by the equipment's own sensors and control system. The sensors transmit information such as the robot arm's position and speed to the control system, which records the corresponding task parameters based on the task schedule. For example, recording the robot arm's movement from a starting point (x1, y1, z1) to the next point (x2, y2, z2) at a set speed and trajectory generates the equipment's motion trajectory data. The current processing task parameters specify the polishing roughness of a specific part of the shell to a specified value, or the specific coordinate position range for the assembly of a specific part.

[0020] Step 112: Determine key processing posture nodes in the equipment motion trajectory data, and deploy multiple image acquisition units under each key processing posture node to generate image acquisition layout strategies for different observation angles.

[0021] During the processing of a robot vacuum's housing, the equipment's motion trajectory data captures the movement of the robotic arm between multiple operating positions. Key processing pose nodes are identified, specifically the robotic arm positions that are crucial for the housing processing. For example, the robotic arm position during operations such as polishing key parts and installing parts is a key processing pose node. The installation pose of one of the housing's panels represents a key processing pose node. To capture images from different observation angles at this pose node, multiple image acquisition units are deployed. Cameras can be installed in various directions around the robotic arm—some above, some to the side, and some diagonally below. This arrangement of cameras constitutes the image acquisition layout strategy. Cameras at different angles capture the robot vacuum's housing in this pose from various viewpoints, providing multi-angle data for accurate subsequent processing analysis.

[0022] Step 113: Based on the image acquisition layout strategy, each image acquisition unit is triggered to perform a synchronous image acquisition operation to generate a synchronous image sequence set associated with multiple pose nodes; wherein the image data in each synchronous image sequence set corresponds to different observation perspectives of the same processing pose node in the time dimension.

[0023] When a predetermined critical processing pose node is reached, the system, following a pre-defined image acquisition layout strategy, issues a unified command to trigger the image acquisition units deployed at each location. This means that cameras at different locations perform synchronized image acquisition operations. For example, for a critical processing pose node on the robot vacuum's housing, five cameras positioned around it simultaneously begin capturing images, each corresponding to a different viewing angle. The resulting images are then categorized by pose node. For example, the images of this assembly pose node form a synchronized image sequence set. Within this set, images captured by different cameras all represent the same part assembly pose from different angles at the same moment (in the temporal dimension). This resulting synchronized image sequence set comprehensively reflects the details of the processing pose, providing comprehensive data for subsequent image analysis.

[0024] Step 114: parsing the target processing object contour data in the current processing task parameters, and removing image frames that do not meet the image coverage constraint from the synchronous image sequence set according to the target processing object contour data to generate the original image sequence.

[0025] In the processing of the shell of a sweeping robot, the target processing object contour data in the current processing task parameters clearly defines the accurate contour shape and size range of each part to be processed of the sweeping robot shell. When screening the original image sequence from the synchronized image sequence set, a comparison of image coverage constraints is performed. For example, some images taken by cameras only show part of the shell of the sweeping robot and do not fully cover the part of interest of the current processing task, that is, image frames that do not meet the image coverage constraints. In this case, these image frames are removed from the synchronized image sequence set, and the remaining images constitute the original image sequence. The original image sequence will become the input data for the image state adaptive analysis processing in the subsequent steps to ensure that the subsequent analysis is based on appropriate and complete image data.

[0026] Step 120: performing image state adaptive analysis processing on the original image sequence to obtain an image recognition feature set for each image region in the original image sequence, wherein the image recognition feature set includes spatial distribution features, image analysis features, and environmental interference features.

[0027] In a scenario where the robot vacuum's shell is the processing object, the original image sequence records the state of the robot vacuum's shell at different key processing positions. For each image, image state adaptive analysis processing is performed. First, each image region is analyzed. Each region has a corresponding series of features defined as an image recognition feature set. Spatial distribution features are characteristics related to the position and shape of the image region in the image; image parsing features are used to analyze the detailed features of the target object of the shell itself; and environmental interference features focus on the impact of the processing environment on the shell image. Each image region is analyzed in this way to form an image recognition feature set for each image region. For example, in one of the original images, for the image region where one of the protruding parts of the robot vacuum's shell is located, multiple pieces of information such as its spatial position, its own contour details, and whether it is subject to environmental interference must be determined.

[0028] In a preferred embodiment, performing image state adaptive analysis on the original image sequence to obtain an image recognition feature set for each image region in the original image sequence includes: Step 121: performing multi-scale spatial segmentation processing on each image frame in the original image sequence, dividing the image frame into multiple candidate regions, and performing edge structure extraction on each candidate region.

[0029] For each frame in the original image sequence, using the robot vacuum shell image as an example, a multi-scale spatial segmentation process is performed. This uses a segmentation method based on a specified algorithm (such as the watershed algorithm). This algorithm analyzes the image's grayscale values ​​to divide the image into distinct regions, from coarse to fine. The image is first roughly divided into several large regions, which are then further subdivided. After this multi-scale segmentation, the robot vacuum shell image is divided into multiple candidate regions. Edge structure extraction is then performed on each candidate region, using an edge detection algorithm (such as the Canny edge detection algorithm) to identify the edges of each candidate region. For example, in the candidate region where a leg component of the robot vacuum shell is located, the Canny algorithm can accurately mark the edge position and direction between the component and the surrounding environment, thereby capturing the edge structure.

[0030] Step 122: Determine the spatial distribution characteristics of the candidate region based on the edge structure extraction result, where the spatial distribution characteristics include the coordinates of the region center point, the region geometric parameters, and the relative distance distribution between the region and the image boundary.

[0031] Based on the edge structure information extracted previously, the spatial distribution characteristics of each candidate area are determined. Taking the candidate area where the battery compartment on the outer shell of the sweeping robot is located as an example, the coordinates of the center point of the area are calculated. The center point coordinates (x0, y0) can be obtained by performing operations on the coordinates of all pixel points in the area (such as finding the average value). In terms of regional geometric parameters, an algorithm is used to calculate the shape parameters of the candidate area, such as length, width, area, perimeter, etc.; the relative distance distribution between the area and the image boundary is determined by measuring the distance from the candidate area to the left and right boundaries and the upper and lower boundaries of the image. These spatial distribution characteristics are important for fully understanding the position and shape of the candidate area in the image, and provide key information for establishing a connection with the positioning status of the processing equipment.

[0032] Step 123: performing image analysis feature extraction on each candidate region, wherein the image analysis features include local deformation features of the target object contour, surface material reflection features, and edge connectivity features with adjacent regions.

[0033] Continue to perform image parsing feature extraction for each candidate area, still taking the battery compartment candidate area as an example. For the local deformation characteristics of the target object contour, by comparing with the pre-set standard battery compartment contour, find the subtle deformation parts of the battery compartment contour in the image, and quantify the degree of these deformations. In terms of surface material reflection characteristics, the characteristics of image pixel values ​​and optical principles are used to analyze the reflection of the surface material of the battery compartment. Because different materials reflect differently, the surface material reflection characteristics of the area are determined by analyzing the changes in pixel values. The edge connectivity characteristics between adjacent areas study the degree of connection and method of the edges between adjacent areas of the battery compartment to determine whether there are gaps or fusion areas. These image parsing features can help understand the details of the target object and its relationship with the surrounding areas, which is critical for the planning of processing operations.

[0034] Step 124: performing environmental interference feature recognition processing on the candidate area, where the environmental interference features include imaging illumination distribution uniformity features, background interference noise features, and blur features of the target object in the image.

[0035] In the image of the robot vacuum shell processing scenario, environmental interference features are identified for each candidate area. Imaging illumination distribution uniformity features are used to check whether the candidate area in the image has uniform illumination. This can be determined by calculating the difference in pixel values ​​at different locations within the area. Large differences indicate uneven illumination. Background interference noise feature identification uses a preset algorithm (such as a Gaussian noise detection algorithm) to determine whether noise interference exists in the candidate area in the image, as well as information such as the distribution and intensity of the noise. The blurred features of the target object in the image are observed. If the target object (such as a component of the robot vacuum shell) is blurred in the image, the direction of the blur (horizontal, vertical, or diagonal) and the degree of blur (such as measured by parameters such as the blur radius) are determined. These environmental interference features are important for correcting image-based processing judgments and avoiding processing errors caused by interference factors.

[0036] Step 125: Associate the spatial distribution features, the image analysis features, and the environmental interference features and store them as an image recognition feature set of the image region.

[0037] After analyzing the spatial distribution characteristics, image parsing features, and environmental interference characteristics of each candidate region, these features are associated and stored to form an image recognition feature set, taking a specific region in the image of the robot vacuum's housing as an example. This can be achieved by using a target data structure (such as a hash table), using the region identifier as an index and associating the three corresponding features as storage content. This image recognition feature set comprehensively covers the region's spatial position, target object details, and environmental interference, providing a complete data foundation for subsequent positioning state mapping processing using these features.

[0038] Step 130: Perform positioning state mapping processing on the image recognition feature set based on the device positioning state data set to generate a spatial positioning drive feature set corresponding to each image area; the spatial positioning drive feature set is used to characterize the spatial trajectory adjustment amount and posture correction direction required for the processing equipment to perform the target processing operation under the corresponding observation posture.

[0039] During the processing of a robot vacuum's shell, the device positioning state data set records the position, posture, and other information of the processing equipment at the time, while the image recognition feature set stores various features of each image area. These two aspects are combined when performing positioning state mapping processing. For example, taking the image area of ​​a robot vacuum's shell where a sensor is to be installed as an example, a mapping relationship is established by setting an algorithm with reference to the device positioning state data set. For example, using a mapping algorithm based on kinematics and vision principles, the spatial trajectory and posture correction direction that the robotic arm needs to adjust when the processing equipment performs the sensor installation operation in that area are derived from the features corresponding to the image area, forming a spatial positioning drive feature set. This set provides clear operating instructions for the processing equipment, ensuring the accurate implementation of the processing operation.

[0040] In an optional embodiment, performing positioning state mapping processing on the image recognition feature set based on the device positioning state data set to generate a spatial positioning drive feature set corresponding to each image area includes: Step 131: extracting device posture parameters corresponding to the current image area from the device positioning state data set, wherein the device posture parameters include the posture coordinates of the end of the robot arm, the tool orientation angle, and the actual contact distance between the processing tool and the target surface.

[0041] Taking the image area corresponding to the camera bracket installation operation on the robot vacuum's housing as an example, the relevant device pose parameters for the processing equipment at that time were retrieved from the device positioning status data set. The corresponding data was found using a preset indexing method (for example, by matching the image acquisition moment with the device status recording time). The robot arm's end pose coordinates record the position of the end arm in three-dimensional space; the tool orientation angle specifies the direction in which the camera bracket installation tool was pointed; and the actual contact distance data between the processing tool and the target surface reflects the distance between the tool and the robot vacuum's housing at that time. These parameters provide input related to the actual processing equipment status for subsequent processing.

[0042] Step 132: Create a spatial relationship mapping network including the spatial distribution features and device posture parameters, and calculate the spatial error vector between the observed posture of the current image area and the expected posture of the device based on the spatial relationship mapping network.

[0043] Optionally, the creating a spatial relationship mapping network including the spatial distribution features and the device posture parameters, and calculating the spatial error vector between the observed posture of the current image area and the expected posture of the device based on the spatial relationship mapping network, includes: Step 1321: Establish a multidimensional mapping topology structure with the coordinates of the center point of the image area and the coordinates of the device end posture as nodes, and use the node connection weights in the multidimensional mapping topology structure as the matching degree between the area geometric parameters and the tool orientation angle.

[0044] Taking the processed image area of ​​one of the decorative parts on the robot vacuum's shell as an example, we begin processing. First, a multi-dimensional mapping topology is established based on the coordinates of the center point of the image area and the coordinates of the device's end pose. This can be a graph-based topology with edges connecting the nodes. In the multi-dimensional mapping topology, the degree of matching between the area's geometric parameters and the tool's orientation angle is used as the node connection weight. For example, if the decorative part is a rectangle and the tool's orientation angle is conducive to processing the rectangle, a higher weight is assigned; otherwise, a lower weight is assigned. This weight setting is based on experience or a trained model to prepare for subsequent error transmission and calculation.

[0045] Step 1322: construct an error transmission link between nodes based on the multi-dimensional mapping topology structure, and transmit the relative distance distribution features in the spatial distribution features to the terminal pose coordinate nodes in sequence through the error transmission link.

[0046] Based on the established multidimensional mapping topology, an error transmission link is constructed, and a shortest path algorithm is used to find the path. Taking the image region containing the decorative component as an example, the relative distance distribution feature within the spatial distribution features describes the distance relationship between the image region and the image boundary. This feature information is then transmitted to the end pose coordinate node through this link. For example, if the relative distance distribution feature indicates that the region is close to one side of the image boundary, this information is transmitted through the link, allowing the end pose coordinate node to understand this spatial relationship and prepare for data flow for the subsequent calculation of spatial error.

[0047] Step 1323: Introduce the environmental interference feature as a link correction factor for each intermediate node in the error conduction link, and update the distribution value of the node connection weight.

[0048] At each intermediate node in the error transmission chain, using the decorative component image region as an example, environmental interference features are introduced. For example, if the illumination in the image region is uneven, this environmental interference feature is used as a parameter to correct the node connection weight distribution. A preset correction algorithm (such as the weighted average correction algorithm) can be used to update the connection weights. The weights connecting each node are recalculated based on the degree and direction of the uneven illumination. For example, the weight corresponding to the direction with high illumination intensity is appropriately reduced, as excessive illumination may affect the processing accuracy judgment.

[0049] Step 1324: Perform back propagation calculation on the spatial error vector based on the updated node connection weights to obtain an initial spatial error vector between the observed pose of each image region and the expected pose of the device.

[0050] Backpropagation calculations are performed using the updated node connection weights to derive the initial spatial error vector. Using a backpropagation algorithm (similar to the backpropagation algorithm in neural networks), the error signal is propagated back from the end pose coordinate node. By calculating and deducing the connection weights of each node, the initial spatial error vector between the observed pose and the desired pose of the device is obtained for each image region. This vector reflects the spatial position and posture deviation from the current observation angle to the desired processing pose.

[0051] Step 133: performing interference weight allocation processing on the environmental interference feature, correcting the spatial error vector according to the interference weight allocation processing result, and generating a corrected spatial error correction value.

[0052] In an optional embodiment, performing interference weight allocation processing on the environmental interference feature, correcting the spatial error vector according to the interference weight allocation processing result, and generating a corrected spatial error correction value includes: Step 1331: Determine the illumination intensity gradient distribution of the image area according to the imaging illumination distribution uniformity feature, and calculate the influence factor of the illumination intensity gradient distribution on the positioning accuracy of the target object contour.

[0053] Taking one of the sensor mounting areas on the robot vacuum's housing as an example, the uniformity of the imaging illumination distribution is analyzed. The gradient distribution of illumination intensity in this area can be determined by calculating the grayscale difference between adjacent pixels. An analytical algorithm (such as a model trained using machine learning) is then used to calculate the impact factor of this gradient distribution on the positioning accuracy of the target sensor mounting area's outline. Larger variations in illumination gradient indicate a higher impact factor, while smaller variations indicate a lower impact factor. This factor is used for subsequent corrections.

[0054] Step 1332: extract the target noise amplitude and target interference distribution pattern from the background interference noise feature, and determine the influence weight of the target noise on the edge positioning accuracy and the influence weight of the target interference on the region segmentation stability.

[0055] The target noise amplitude and distribution pattern are extracted from the background interference noise characteristics. Taking the sensor installation area as an example, a noise detection and analysis algorithm is used to determine the noise amplitude and distribution pattern (e.g., random distribution or concentration in a specific pixel area). Based on these characteristics and processing standards, the weight of the target noise's impact on regional edge positioning accuracy and regional segmentation stability is determined. If the noise amplitude is large and concentrated in the edge area, the impact on edge positioning accuracy is greater.

[0056] Step 1333: Analyze the blur direction and blur length in the blur feature, combine the actual motion speed in the device motion trajectory data, and calculate the error contribution weight of the blur to the image analysis feature.

[0057] For the blur characteristics of the image in the sensor installation area, measure the blur direction (horizontal, vertical, or diagonal) and blur length (measured using one of the following quantitative indicators). Combined with the actual motion speed in the device's motion trajectory data, a calculation method (such as one based on a combination of fuzzy mathematics and kinematics) is used to calculate the error contribution weight of the blur to the image analysis features. If the blur direction is related to the device's motion direction and the blur length is large, the error contribution weight is increased accordingly.

[0058] Step 1334: Standardize the positioning accuracy influencing factor, the influence weight of the target noise on the edge positioning accuracy, the influence weight of the target interference on the regional segmentation stability, and the error contribution weight to generate a comprehensive interference weight distribution matrix.

[0059] The previously calculated factors affecting positioning accuracy, the weights of target noise impacting edge positioning accuracy, the weights of target interference impacting region segmentation stability, and the weights of error contributions are combined and normalized. A preset normalization algorithm (such as the Min-Max normalization algorithm) is used to normalize each factor and weight so that their values ​​fall within a preset range, such as [0, 1]. The purpose of normalization is to allow different weight types to be compared and calculated on the same scale. By applying these values ​​to the normalization algorithm, a comprehensive interference weight distribution matrix is ​​generated. This matrix contains the comprehensive weights of different environmental interference factors related to image analysis and subsequent processing and positioning. It comprehensively reflects the degree of environmental interference's impact on processing and the relative relationships between these factors. For example, the matrix allows for a clearer picture of whether illumination interference or noise interference has a greater impact on the positioning of the current image region.

[0060] Step 1335: Generate a weighted correction operator based on the comprehensive interference weight distribution matrix, apply the weighted correction operator to the initial spatial error vector, and obtain the corrected spatial error correction amount.

[0061] Based on the previously generated comprehensive interference weight distribution matrix, a weighted correction operator is generated according to the corresponding algorithm. This weighted correction operator is an operation matrix or function that contains information about the weights of various interference factors. This weighted correction operator is applied to the previously calculated initial spatial error vector, which can be calculated through matrix multiplication or function application. After the above calculations, the initial spatial error vector is adjusted accordingly based on the weights of different environmental interference factors, ultimately resulting in a corrected spatial error correction value. The corrected spatial error correction value more accurately reflects the actual error between the observed pose of the image area and the expected pose of the device, taking into account environmental interference factors. For example, it more accurately provides the spatial position and attitude deviation values ​​that need to be adjusted for the robot arm after compensating for the influence of environmental interference, providing more reliable data for the subsequent generation of accurate spatial positioning drive features.

[0062] Step 134: Determine the coverage adjustment requirements of the processing tool on the target surface based on the local deformation features of the target object contour in the image analysis features, and generate the spatial positioning drive feature set in combination with the corrected spatial error correction amount; wherein, the spatial positioning drive feature set includes the displacement compensation amount of the end of the robot arm, the tool posture rotation angle and the adjustment coefficient of the processing pressure.

[0063] Taking the radiator mounting area on a robot vacuum's housing as an example, the local deformation characteristics of the radiator's contour are analyzed based on image analysis features. If deformations such as localized protrusions or depressions are detected, corresponding algorithms and rules are used to determine the required coverage adjustment of the machining tool on the target surface when installing the radiator. For example, if one corner of the radiator is deformed, the installation tool's coverage pattern and area of ​​that corner may need to be adjusted. This information is then combined with the corrected spatial error correction to generate a spatial positioning drive feature set. The displacement compensation of the robot arm's end is determined based on the spatial error correction and the required coverage adjustment to ensure the robot arm accurately reaches the appropriate position. The tool's posture rotation angle is adjusted based on the tool's compatibility with the deformed contour and spatial deviations. The machining pressure adjustment coefficient takes into account factors such as the housing material and installation requirements to ensure that the appropriate pressure is applied during radiator installation. Together, these factors constitute the spatial positioning drive feature set, providing specific drive parameters for the machining operation.

[0064] In detail, the generating of the spatial positioning driving feature set by combining the corrected spatial error correction amount includes: Step 1341: Obtain positioning accuracy constraints in the target processing task parameters, where the positioning accuracy constraints include an upper tolerance limit for axial positioning error, an angle deviation threshold, and an allowable fluctuation range for contact pressure.

[0065] In the robot vacuum shell processing task, for the operation of installing one of the set parts (such as the motherboard fixing screws), the positioning accuracy constraints in the target processing task parameters are obtained. The upper limit of the axial positioning error tolerance sets the maximum error value allowed when the robot arm moves along the set coordinate axis. The angular deviation threshold specifies the maximum angle at which the processing tool can deviate from the ideal direction in posture. The allowable fluctuation range of contact pressure clarifies the numerical range in which the pressure of the processing tool can fluctuate when it contacts the robot vacuum shell. These constraints are determined based on the design requirements and processing technology standards of the robot vacuum shell. For example, in order to ensure that the motherboard can work normally after installation, the position error of the screw installation must be controlled within a very small range, and the angular deviation cannot affect the tightening effect of the screw. Excessive contact pressure may damage the shell, and too small a contact pressure may cause the screw to be loosely fixed.

[0066] Step 1342: Calculate the proportional coefficient between the corrected spatial error correction amount and the upper tolerance limit of the axial positioning error, and determine the adaptive adjustment step size of the displacement compensation amount of the robot end according to the proportional coefficient.

[0067] Optionally, the corrected spatial error correction is compared with the upper tolerance limit for axial positioning error. Taking the error and constraints corresponding to the image area where the motherboard's fixing screws are installed as an example, the axial displacement error component in the spatial error correction is divided by the upper tolerance limit for the axial positioning error to obtain a proportionality factor. This proportionality factor reflects the relative relationship between the current error and the acceptable error. Based on this proportionality factor, a preset adaptive adjustment algorithm (e.g., one based on a linear relationship) is used to determine the adaptive adjustment step size for the displacement compensation of the end arm. A large proportionality factor indicates that the error is close to the upper tolerance limit, and the adjustment step size is relatively large. Conversely, a small proportionality factor results in a small adjustment step size. This adaptive adjustment ensures that the robot arm neither over- nor under-adjusts when compensating for errors.

[0068] Step 1343: performing dead-zone filtering on the tool posture rotation angle according to the angle deviation threshold, setting the rotation angle component smaller than the angle deviation threshold to zero, and filtering out small angle jitter interference.

[0069] When processing the rotation angle of the machining tool, taking the motherboard screw installation tool as an example, dead-zone filtering is performed with reference to the angle deviation threshold. The angle deviation threshold sets an acceptable minimum angle variation range. Components of the tool's rotation angle that are smaller than this threshold are detected. These slight angle variations may be caused by slight jitter or noise interference in the equipment and will not have a substantial impact on the machining process. Based on the dead-zone filtering algorithm, these rotation angle components that are smaller than the angle deviation threshold are set to zero, thereby filtering out slight angle jitter interference. This allows tool posture angle adjustments to be more based on the large angle deviations that actually need to be adjusted, improving machining accuracy and stability and avoiding time consumption and equipment wear caused by unnecessary slight angle adjustments.

[0070] Step 1344: applying a boundary constraint of the allowable fluctuation range of the contact pressure to the adjustment coefficient of the processing pressure, and clipping the adjustment coefficient to the nearest boundary value when it exceeds the boundary constraint.

[0071] Regarding the adjustment coefficient for processing pressure, taking the pressure adjustment during motherboard screw installation as an example, the boundary constraint of the allowable contact pressure fluctuation range is considered. If the calculated processing pressure adjustment coefficient exceeds the allowable fluctuation range, the adjustment coefficient is adjusted to the nearest boundary value according to preset clipping rules (such as directly comparing the size with the boundary value for clipping). For example, if the adjustment coefficient is greater than the upper limit of the allowable fluctuation range, it is clipped to the upper limit; if it is less than the lower limit, it is adjusted to the lower limit. This ensures that the processing pressure remains within a reasonable range, avoiding problems such as excessive pressure damaging the robot vacuum's casing or insufficient pressure causing insecure installation.

[0072] Step 1345: Generate a spatial positioning drive feature set that meets the positioning accuracy constraint condition based on the adaptive adjustment step size, the tool posture rotation angle after the dead zone filtering processing, and the adjustment coefficient after clipping.

[0073] The adaptive adjustment step size, the tool rotation angle after dead-zone filtering, and the trimmed adjustment coefficient are integrated together. Taking motherboard screw installation as an example, these parameters, derived from the robot arm's movement position, tool orientation, and processing pressure, together constitute a set of spatial positioning drive features that meet positioning accuracy constraints. The parameter values ​​in this set are derived through a series of calculations and corrections. They work together to ensure that the processing equipment can perform the corresponding processing operations on the robot vacuum housing with high precision. For example, the robot arm accurately moves to the screw installation position, the tool tightens the screw with the appropriate posture, and the applied pressure meets the requirements.

[0074] Step 140: Generate an adaptive positioning control instruction based on the spatial positioning drive feature set and transmit it to the execution end of the processing equipment, instructing the processing equipment to perform an adaptive posture positioning control operation for the current image area.

[0075] After calculating and generating the spatial positioning drive feature set, these features are used to generate adaptive positioning control instructions, exemplified by drilling a hole in a designated area of ​​a robot vacuum's housing. These instructions are encoded and converted based on information from the spatial positioning drive feature set, such as the manipulator's end-of-arm displacement compensation, tool rotation angle, and machining pressure adjustment coefficient, into a command format that the machining equipment can recognize and execute. These instructions are then packaged into data packets using a pre-defined communication protocol to ensure data accuracy and integrity during transmission. These adaptive positioning control instruction packets are then transmitted to the machining equipment's execution terminal via a real-time bus. Upon receiving the data packets, the execution terminal parses the instructions and adjusts the position of the manipulator and machining tool accordingly, achieving adaptive position and positioning control. For example, the manipulator arm carrying the drilling tool accurately moves to a specified position and position, allowing the drilling operation to proceed with appropriate pressure.

[0076] As an implementation method, the generating of the adaptive positioning control instruction according to the spatial positioning drive feature set and transmitting the instruction to the execution end of the processing equipment to instruct the processing equipment to perform the adaptive posture positioning control operation for the current image area includes: Step 141: Generate a path planning intermediate node sequence according to the displacement compensation amount of the end of the robot arm, and insert a smooth transition trajectory into the path planning intermediate node sequence to eliminate the sudden stop jitter during the movement of the robot arm.

[0077] The operation of installing decorative strips on the outer shell of a sweeping robot is used as a detailed explanation. Based on the displacement compensation of the end of the robotic arm, a path planning algorithm (such as the A* algorithm and other search and optimization-based route algorithms) is used to generate a path planning intermediate node sequence. This sequence determines a series of intermediate position points in the process of the robotic arm moving from the current position to the target position for installing the decorative strip. Considering the smoothness of the robotic arm's movement, inserting a smooth transition trajectory into this intermediate node sequence can be achieved through a spline interpolation algorithm. This algorithm generates a smooth curve based on the existing node position information, allowing the robotic arm to move smoothly along this curve, avoiding sudden stops and jitters during movement, and ensuring the accuracy and stability of the decorative strip installation. For example, it can prevent the decorative strip from fitting loosely to the outer shell due to jitters during the installation process.

[0078] Step 142: Convert the tool posture rotation angle into a quaternion representation and associate it with each node in the path planning intermediate node sequence to generate a continuous posture control instruction stream for the end effector of the robot arm.

[0079] For the posture rotation angle of the tool for installing decorative strips, taking the angle corresponding to the current processing state as an example, it is converted into a quaternion representation. Quaternion is a mathematical tool for representing rotation, which can more conveniently perform posture calculation and transformation. The converted quaternion is associated with each node in the intermediate node sequence of path planning. Each node contains the posture information that the tool should have when the robot arm reaches that point. In this way, a continuous posture control instruction stream of the end effector of the robot arm is generated, so that the robot arm can adjust the posture of the tool in real time according to the posture information corresponding to each node while moving along the planned path, ensuring that the tool always maintains a suitable posture and contacts the shell surface during the installation of the decorative strip, achieving a good installation effect.

[0080] Step 143: performing time-domain discretization processing on the adjustment coefficient of the machining pressure to obtain a pressure adjustment gradient curve for each time segment, and coupling the pressure adjustment gradient curve with a servo motor control signal of the machining tool.

[0081] Taking the example of the need for appropriate pressure when installing decorative strips, the adjustment coefficient for the processing pressure is discretized in the time domain. Using a preset discretization algorithm (such as an equally spaced sampling algorithm), the time axis of the entire processing process is divided into multiple time segments. The pressure adjustment coefficient is determined for each segment, resulting in a pressure adjustment gradient curve for each time segment. This curve describes the temporal variation of the processing pressure. This pressure adjustment gradient curve is then coupled with the control signal of the servo motor of the processing tool. The servo motor controls the movement and force of the processing tool. Using a coupling algorithm (for example, correlating pressure with motor current or speed), the servo motor adjusts its output according to the pressure adjustment gradient curve, thereby varying the pressure applied by the processing tool. For example, when the decorative strip first contacts the casing, the pressure is gradually increased. After reaching a certain level, a stable pressure is maintained during installation to ensure a secure installation without damaging the casing.

[0082] Step 144: Merge the continuous posture control instruction stream and the coupled servo motor control signal into a unified time synchronization control instruction set, and encapsulate it into an adaptive positioning control instruction data packet according to the processing equipment communication protocol.

[0083] The previously generated continuous position and posture control command stream for the robot's end effector and the servo motor control signals coupled with pressure regulation information are merged. Using a pre-set merging algorithm (e.g., alignment and integration based on timestamps), a unified, time-synchronized control command set is formed. This command set encompasses all control information, including the robot's position and posture, as well as the tool pressure, and is synchronized in time to ensure coordinated operation. This command set is then encapsulated according to the processing equipment communication protocol (e.g., a pre-set industry standard communication protocol that specifies data format and transmission rules) and converted into adaptive positioning control command data packets, enabling accurate and reliable data transmission across the processing equipment network.

[0084] Step 145: Transmit the adaptive positioning control instruction data packet to the execution end of the processing equipment via the real-time bus, instructing the execution end to parse the data packet and drive the robot arm and the processing tool to perform a collaborative positioning operation.

[0085] Optionally, the packaged adaptive positioning control instruction data packet is sent out through the real-time bus, which ensures that the data can be transmitted quickly and accurately to the execution end of the processing equipment. Taking the installation device of the decorative strip of the sweeping robot shell as an example, after receiving the data packet, the execution end parses it according to the parsing rules of the communication protocol. The control information such as the position, posture and pressure of the processing tool of the robot arm is extracted from the data packet, and this information is sent to the corresponding control module to drive the robot arm and the processing tool to perform collaborative positioning operations. For example, the robot arm accurately moves to the installation position of the decorative strip and adjusts its posture according to the position and posture information. At the same time, the processing tool firmly installs the decorative strip on the shell with appropriate pressure according to the pressure information.

[0086] Preferably, after the indication processing device performs the adaptive posture positioning control operation on the current image area, it further includes: Step 210: Capture the actual posture feedback data of the execution end of the processing equipment and the quality inspection image data of the processing surface in real time.

[0087] After completing adaptive position positioning control operations on a specific area of ​​the robot vacuum housing (such as the camera mounting area), the system begins capturing relevant data in real time. Actual position feedback data from the processing equipment's execution end is collected in real time by position and attitude sensors installed on the end of the robotic arm and the processing tool. These sensors provide feedback on information such as the actual position coordinates reached by the robotic arm and the tool's final attitude angle. For example, the final position returned by the robotic arm end is (x3, y3, z3), and the tool's attitude rotation angle is (α, β, γ). Simultaneously, a dedicated quality inspection camera captures real-time image data of the processed surface. This camera captures the camera mounting area of ​​the processed robot vacuum housing from a set angle, capturing surface quality information such as surface scratches and alignment of installed components, providing a data foundation for subsequent analysis.

[0088] Step 220: Perform deviation analysis on the actual posture feedback data and the expected parameters in the spatial positioning drive feature set, and trigger re-acquisition of the multi-source image data set to perform a new round of adaptive positioning control when the deviation exceeds an adjustment threshold.

[0089] The actual pose feedback data is compared and analyzed with the expected parameters from the spatial positioning drive feature set. For example, for operations in the camera installation area, the actual robot arm position coordinates (x3, y3, z3) are compared with the ideal position coordinates corresponding to the expected robot end displacement compensation in the spatial positioning drive feature set to calculate the position deviation. Similarly, the actual tool rotation angles (α, β, γ) are compared with the expected tool rotation angles. These deviations are quantitatively analyzed. If the deviation exceeds a pre-set adjustment threshold, it indicates that the current machining operation may have accuracy issues. This triggers the re-acquisition of the multi-source image data set, re-collecting images from various angles and recording the device positioning status data. This is then used to execute a new round of adaptive positioning control to correct the deviation and ensure the accuracy of subsequent machining operations.

[0090] Step 230: Perform surface defect recognition processing on the quality inspection image data. When defects exceeding the allowable standard are detected on the processed surface, parameter tuning learning is performed based on the corresponding spatial positioning drive feature set based on the defect distribution characteristics, and the positioning control strategy in subsequent processing tasks is updated.

[0091] It is understandable that specialized surface defect recognition algorithms (such as deep learning-based image recognition algorithms that use pre-trained models to identify defect features in images) are applied to quality inspection image data. Taking the image of the camera installation area of ​​a sweeping robot's housing as an example, if surface defects such as scratches or uneven installation are detected, and these defects exceed pre-set allowable standards, the previously generated spatial positioning drive feature set is traced back based on the distribution characteristics of the defects, such as the location of the scratches and the extent of the uneven installation area. The system analyzes which parameter settings may have caused these defects, such as whether the displacement compensation of the end of the robot arm is inaccurate or the tool posture rotation angle is deviated. This analysis allows for parameter tuning and learning, adjusting relevant parameters and updating the positioning control strategy for subsequent processing tasks to prevent similar defects from recurring.

[0092] Step 240: The actual posture feedback data is associated with the quality inspection image data and stored in a processing knowledge base as an optimized training sample for generating adaptive positioning control instructions in the next batch of processing tasks.

[0093] The actual position feedback data and quality inspection image data are associated. For example, the actual position and posture information of the robot arm is associated with the quality inspection image captured by the corresponding camera using pre-set identifiers (such as the processing task ID and timestamp). This associated data is then stored in the processing process knowledge base, a large-capacity data storage system used to store various useful information in the processing process. This data serves as optimized training samples for generating adaptive positioning control instructions in the next batch of processing tasks. By learning and analyzing these samples (for example, using model training algorithms in machine learning), the algorithm and parameter settings can be optimized, making the subsequently generated adaptive positioning control instructions more accurate and reliable, and continuously improving processing quality and efficiency.

[0094] As a non-limiting embodiment, after generating the adaptive positioning control instruction according to the spatial positioning drive feature set and transmitting it to the processing equipment execution end, the method further includes: monitoring the posture response data stream of the processing equipment execution end in real time, and extracting the actual displacement compensation amount, the actual posture rotation angle, and the pressure adjustment instantaneous value from the posture response data stream; Performing differential calculation on the actual displacement compensation amount and the manipulator end displacement compensation amount in the spatial positioning drive feature set to generate an axial displacement residual sequence and an attitude angle deviation matrix; Constructing a spatiotemporal distribution map of posture errors based on the axial displacement residual sequence, marking abnormal response nodes exceeding a preset error band in the spatiotemporal distribution map and associating them with corresponding image region identifiers; According to the abnormal response node, a local image re-acquisition instruction is triggered, and a high-resolution scanning imaging module is started to perform sub-pixel image acquisition for the associated image area under the current observation posture of the processing equipment; Performing multi-level feature comparison processing on the sub-pixel image, calculating the real-time offset of each feature dimension in the image recognition feature set and reversely correcting the displacement compensation parameters in the spatial positioning drive feature set; The corrected displacement compensation parameter is injected into the data encapsulation queue of the adaptive positioning control instruction, overwriting the original control parameter and driving the processing equipment to perform a cyclic adjustment process.

[0095] In this embodiment, the posture response data stream emitted by the execution end of the processing equipment is monitored in real time. Taking the processing equipment at the assembly point of a part of the sweeping robot shell as an example, the actual displacement compensation, actual posture rotation angle and pressure adjustment instantaneous value are extracted from it. The actual displacement compensation reflects the difference between the actual movement distance of the robot arm and the set value, the actual posture rotation angle is the final posture change of the tool, and the pressure adjustment instantaneous value represents the pressure applied by the processing tool at the current moment. The actual displacement compensation is calculated by differential calculation with the displacement compensation of the end of the robot arm in the spatial positioning drive feature set to obtain the axial displacement residual sequence in each coordinate axis direction and the posture angle deviation matrix of the tool posture in each direction. The axial displacement residual sequence shows the change of the deviation between the actual position and the expected position of the robot arm in each axial direction over time; the posture angle deviation matrix comprehensively reflects the posture deviation of the tool in different directions.

[0096] Based on the axial displacement residual sequence, a spatiotemporal distribution diagram of the posture error is constructed. This diagram uses time as the horizontal axis and spatial position (such as the three axial directions of the robot arm in the Cartesian coordinate system) as the vertical axis. The axial displacement residual at each moment is plotted on the diagram, forming a distribution map that reflects the changes in posture error over time and space. In this distribution diagram, a preset error band is set, which is an acceptable error range determined based on the machining accuracy requirements. Abnormal response nodes that exceed the preset error band are marked. These nodes represent the time and location of large posture deviations during the machining process. These abnormal response nodes are associated with the corresponding image area identifiers to clearly identify the machining area corresponding to the deviation.

[0097] Once an abnormal response node is found, the system will trigger a local image re-acquisition instruction based on it. Under the current observation posture of the processing equipment, the high-resolution scanning imaging module is started to perform sub-pixel image acquisition for the image area associated with the abnormal response node. Sub-pixel images have a higher resolution than ordinary pixels and can capture more subtle image features, providing richer data for subsequent precise analysis. For example, for the posture deviation of a key component installation area on the outer shell of a sweeping robot, a clearer image of the area can be captured through the high-resolution scanning imaging module, accurately presenting information such as the surface condition of the area, details of the parts installation position, etc.

[0098] Multi-level feature comparison processing is performed on the acquired sub-pixel image. Feature analysis is performed from different levels and angles of the image, for example, first analyzing the overall image contour features and then delving deeper into local details such as edges and textures. These features are then compared with the original image recognition feature set, and the real-time offset of each feature dimension in the image recognition feature set is calculated. For example, the position and morphology of part edge features in the current image and the original feature set are compared to calculate the offset of various dimensions such as spatial distribution features and image analysis features.

[0099] Based on the calculated real-time offset, the displacement compensation parameters in the spatial positioning drive feature set are reversed. By analyzing the underlying relationship between feature offsets and displacement compensation (for example, by establishing a model-based mapping relationship), the displacement compensation parameters are adjusted to better meet actual processing requirements. For example, if the part's position in the image is found to be offset to the right by a certain distance than expected, the displacement compensation of the end arm in the spatial positioning drive feature set is increased accordingly.

[0100] Finally, the corrected displacement compensation parameters are injected into the data encapsulation queue of the adaptive positioning control instructions, overwriting the original control parameters. After receiving the updated instructions, the processing equipment performs a cyclic adjustment process. The robotic arm readjusts its position based on the new displacement compensation to correct any previous posture deviations, ensuring that the processing task can continue under more accurate parameter control and continuously improving processing accuracy. The entire process achieves dynamic optimization and adjustment of the processing equipment's posture through a series of operations including real-time monitoring, data analysis, image recapture, feature comparison, and parameter correction.

[0101] As a non-limiting embodiment, after generating the adaptive positioning control instruction according to the spatial positioning drive feature set and transmitting it to the execution end of the processing equipment, the method further includes: Analyzing the tool posture rotation angle component in the spatial positioning drive feature set, and decomposing the rotation angle into three orthogonal axial rotation components around the tool coordinate system in a kinematic model of the machining equipment; Establish a geometric mapping relationship between the rotation component and the normal vector of the machining surface, and calculate the equivalent contact pressure distribution gradient of each rotation component according to the actual contact surface curvature radius of the current machining tool; monitoring a data stream of a six-dimensional force sensor at the end of the machining tool, and extracting an actual contact pressure component corresponding to the equivalent contact pressure distribution gradient in the data stream; Constructing an error field between the pressure distribution gradient and the actual contact pressure component, and when the amplitude of the error field exceeds a material deformation threshold, generating a contact pressure compensation vector and superimposing it on the adjustment coefficient in the spatial positioning drive feature set; Performing kinematic inverse decoupling calculation on the updated adjustment coefficient and the tool posture rotation angle to generate a spatial trajectory correction parameter including a pressure-posture coupling compensation term; The motion interpolation node sequence of the end effector of the robot arm is reconstructed based on the spatial trajectory correction parameters, and pressure-adaptive velocity-acceleration hybrid constraints are inserted between adjacent interpolation nodes to achieve constant force processing control.

[0102] Taking the polishing process of a robot vacuum's housing as an example, we first analyze the tool's posture rotation angle components within the spatial positioning drive feature set. The polishing tool's posture rotation angle is a composite angle value. Within the kinematic model of the machining equipment, this rotation angle is decomposed into rotational components about the three orthogonal axes of the tool coordinate system (typically the X, Y, and Z axes). This facilitates more detailed analysis and control of the tool's posture changes. Each rotational component corresponds to a rotational motion in a different direction. For example, rotation about the X axis controls the tool's pitch angle, rotation about the Y axis controls the roll angle, and rotation about the Z axis controls the yaw angle. This decomposition transforms the complex overall rotation into three independent and quantifiable rotational motions.

[0103] Next, a geometric mapping relationship between the rotational component and the normal vector of the machining surface is established. The normal vector of the machining surface is perpendicular to the machining surface and reflects the directional characteristics of the surface. Combined with the actual contact surface curvature radius of the current machining tool, the equivalent contact pressure distribution gradient of each rotational component is calculated using geometric and physical principles. For example, if the grinding tool contacts the curved surface of the sweeping robot shell at one moment, the smaller curvature radius of the contact surface means that the pressure distribution will be more concentrated. According to the relationship between the rotational component and the surface normal vector, the equivalent contact pressure distribution changes in each direction at different rotation angles can be calculated. This pressure distribution gradient information is very important for accurately controlling the machining pressure.

[0104] At the same time, the data stream from the six-dimensional force sensor at the end of the processing tool is monitored in real time. The six-dimensional force sensor can measure force and torque information in multiple directions, from which the actual contact pressure component corresponding to the equivalent contact pressure distribution gradient is extracted. For example, during the polishing process, the sensor provides real-time feedback of pressure data in various directions. By analyzing and extracting this data, the actual pressure value corresponding to the previously calculated equivalent contact pressure distribution gradient is obtained, thereby understanding the actual pressure applied to the vacuum robot shell during the current processing process.

[0105] Based on the extracted actual contact pressure components and the calculated pressure distribution gradient, an error field is constructed between the pressure distribution gradient and the actual contact pressure components. This error field quantifies the difference between the actual contact pressure and the theoretically expected pressure. The difference between the actual contact pressure component and the pressure distribution gradient is calculated, and the amplitude of the error field is determined using a preset algorithm (such as the sum of squares of the differences). When this amplitude exceeds the material deformation threshold (a pre-set value based on the material properties of the robot vacuum housing, exceeding which unacceptable material deformation may occur), it indicates that the current processing pressure may be adversely affecting the material. A contact pressure compensation vector is then generated. The magnitude and direction of this vector are determined based on the error field to compensate for the difference between the actual pressure and the ideal pressure. This compensation vector is then added to the adjustment coefficient in the spatial positioning drive feature set to modify the processing pressure adjustment coefficient and adjust the subsequent processing pressure.

[0106] The updated adjustment coefficients are then coupled with the tool's posture rotation angle for an inverse kinematic solution. Inverse kinematics involves inferring the required motion parameters for each joint of the robot arm based on the known tool end position and posture. During this process, the adjustment coefficients are incorporated into the inverse kinematic solution to account for the effects of pressure on tool motion and posture. This coupled calculation generates spatial trajectory correction parameters that include pressure-posture coupling compensation terms. These parameters account not only for tool position and posture adjustments but also for the effects of machining pressure, comprehensively reflecting the relationship between pressure and posture during machining and their impact on the machining result.

[0107] Finally, the motion interpolation node sequence of the end effector of the robot arm is reconstructed based on these spatial trajectory correction parameters. The motion interpolation node sequence is a series of discrete points on the motion path of the robot arm. The motion trajectory of the robot arm can be planned by controlling these nodes. On the basis of the original node sequence, adjustments and reconstruction are performed according to the spatial trajectory correction parameters to make the motion trajectory of the robot arm more in line with actual processing requirements. At the same time, pressure-adaptive speed-acceleration hybrid constraints are inserted between adjacent interpolation nodes. These constraints dynamically adjust the movement speed and acceleration of the robot arm according to the pressure conditions during the processing, ensuring that the pressure applied to the outer shell of the sweeping robot remains within a relatively stable range during the entire processing process, realizing constant force processing control, thereby ensuring the consistency of the polishing effect, and avoiding the problem of damage to the outer shell material due to excessive pressure, or insufficient polishing due to insufficient pressure, thereby improving the processing quality and efficiency, and ensuring the stability and reliability of the product.

[0108] It can be understood that when implementing the above-mentioned technical solutions of the embodiments of the present invention, in the adaptive analysis and processing of image states, the OpenCV library can be used to implement the watershed algorithm and Canny edge detection, and the segmentation and edge extraction processes can be fully refined by presetting multi-scale segmentation parameters (such as region merging threshold) and edge detection operators (such as Gaussian filter kernel size); for spatial relationship mapping networks, the convolutional neural network structure can be used for reference, and the back propagation principle combined with the kinematic model can be adopted to clarify the positioning error calculation by defining the update rules of the node connection weights (such as the gradient descent method) to avoid the ambiguity of the network topology.

[0109] When processing environmental interference features, illumination uniformity analysis is performed based on standard image processing methods (such as histogram equalization). A Min-Max normalization algorithm is used to pre-set a normalization range (e.g., [0, 1]) for unified weight calculation, ensuring reproducible interference weight distribution. For dimensional processing, all spatial coordinates, angles, and pressure parameters are uniformly calibrated according to the International System of Units (SI) (e.g., millimeters, radians, and Pascals). For example, robot arm displacement compensation is defined in millimeters, tool posture angles are converted to radians, and machining pressure adjustment coefficients are linked to force sensor readings (in Newtons), eliminating unit ambiguity and ensuring data coordination.

[0110] During the control instruction generation stage, the path planning module based on the Robot Operating System (ROS) (such as the A* algorithm) achieves node sequence smoothing, the posture rotation is processed through the quaternion conversion library, and the servo control protocol (such as EtherCAT) is used to couple the pressure gradient curve to ensure the real-time encapsulation and transmission of instructions.

[0111] In addition, for surface defect identification and parameter tuning, pre-trained deep learning models (such as YOLO) can be integrated for defect detection, and the positioning strategy can be optimized through historical data playback, so that the technical solution of the embodiment of the present invention can achieve high-precision adaptive control in the processing scenario.

[0112] The embodiment of the present invention integrates multi-dimensional data information of the processing scene by acquiring a multi-source image data set covering the original image sequence and the equipment positioning state data set of the target processing scene, so that subsequent analysis is based on a comprehensive data set; then, the original image sequence is adaptively analyzed and processed to obtain an image recognition feature set including spatial distribution, image analysis and environmental interference features, which can deeply analyze the properties of the image region from different key dimensions to accurately characterize the unique characteristics of each image region; then, the image recognition feature set is subjected to positioning state mapping processing based on the equipment positioning state data set to generate a spatial positioning drive feature set. This process realizes the effective mapping from image features to the operation adjustments required for the processing equipment, providing a key basis for the precise control of the processing equipment; finally, an adaptive positioning control instruction is generated based on the set and transmitted to the execution end of the processing equipment, so that the processing equipment can adaptively adjust the posture for different image areas, greatly improving the flexibility of the processing equipment in dealing with complex processing scenarios, ensuring the precise execution of target processing operations under a variety of observation postures and environmental conditions, improving processing accuracy and quality, reducing the need for manual intervention, and optimizing the overall processing process.

[0113] Further, Figure 2 The structure block diagram of the processing positioning system 300 is shown, which includes: a memory 310 for storing program instructions and data; a processor 320 for coupling with the memory 310 and executing the instructions in the memory 310 to implement the above method.

[0114] Furthermore, a computer storage medium is provided, comprising instructions, which implement the above method when executed on a processor.

[0115] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A processing positioning method based on adaptive image recognition, characterized in that: include: Acquire a multi-source image data set of a target processing scene, wherein the multi-source image data set includes a sequence of original images acquired at different observation postures of the processing equipment and a corresponding equipment positioning state data set; Performing image state adaptive analysis processing on the original image sequence to obtain an image recognition feature set for each image region in the original image sequence, wherein the image recognition feature set includes spatial distribution features, image analysis features, and environmental interference features; Performing positioning state mapping processing on the image recognition feature set based on the device positioning state data set to generate a spatial positioning drive feature set corresponding to each image area; The spatial positioning drive feature set is used to characterize the spatial trajectory adjustment amount and posture correction direction required for the processing equipment to perform the target processing operation under the corresponding observation posture; An adaptive positioning control instruction is generated according to the spatial positioning drive feature set and transmitted to the execution end of the processing equipment, instructing the processing equipment to perform an adaptive posture positioning control operation for the current image area.

2. The processing positioning method of adaptive image recognition according to claim 1, characterized in that: Obtain the original image sequence from the multi-source image data set of the target processing scene, including: Acquire a real-time working status data set of a processing device, and analyze the device motion trajectory data and current processing task parameters in the real-time working status data set; Determining key processing pose nodes in the motion trajectory data of the device, and deploying multiple image acquisition units under each key processing pose node to generate image acquisition layout strategies with different observation angles; triggering each image acquisition unit to perform a synchronized image acquisition operation based on the image acquisition layout strategy to generate a synchronized image sequence set associated with a plurality of pose nodes; wherein the image data in each synchronized image sequence set corresponds to different observation angles of the same processing pose node in a temporal dimension; Analyze the target processing object contour data in the current processing task parameters, and remove image frames that do not meet the image coverage constraint from the synchronous image sequence set according to the target processing object contour data to generate the original image sequence.

3. The processing positioning method of adaptive image recognition according to claim 2, characterized in that: The performing image state adaptive analysis processing on the original image sequence to obtain an image recognition feature set for each image region in the original image sequence includes: Performing multi-scale spatial segmentation processing on each image frame in the original image sequence, dividing the image frame into a plurality of candidate regions, and performing edge structure extraction on each candidate region; Determine the spatial distribution characteristics of the candidate region based on the edge structure extraction result, wherein the spatial distribution characteristics include the coordinates of the region center point, the region geometric parameters, and the relative distance distribution between the region and the image boundary; Perform image parsing feature extraction on each candidate region, wherein the image parsing features include local deformation features of the target object contour, surface material reflection features, and edge connectivity features with adjacent regions; Performing environmental interference feature recognition processing on the candidate area, wherein the environmental interference features include imaging illumination distribution uniformity features, background interference noise features, and blur features of the target object in the image; The spatial distribution features, the image analysis features and the environmental interference features are associated and stored as an image recognition feature set of the image area.

4. The processing positioning method of adaptive image recognition according to claim 3, characterized in that: The performing positioning state mapping processing on the image recognition feature set based on the device positioning state data set to generate a spatial positioning drive feature set corresponding to each image area includes: Extracting device posture parameters corresponding to the current image area from the device positioning state data set, the device posture parameters including the posture coordinates of the end of the robot arm, the tool orientation angle, and the actual contact distance between the processing tool and the target surface; Creating a spatial relationship mapping network including the spatial distribution features and device posture parameters, and calculating a spatial error vector between an observed posture of a current image region and an expected posture of the device based on the spatial relationship mapping network; performing interference weight distribution processing on the environmental interference feature, and correcting the spatial error vector according to the interference weight distribution processing result to generate a corrected spatial error correction value; The coverage adjustment requirements of the processing tool on the target surface are determined based on the local deformation features of the target object contour in the image analysis features, and the spatial positioning drive feature set is generated in combination with the corrected spatial error correction amount; wherein, the spatial positioning drive feature set includes the displacement compensation amount of the robot end, the tool posture rotation angle and the adjustment coefficient of the processing pressure.

5. The processing positioning method of adaptive image recognition according to claim 4, characterized in that: The creating of a spatial relationship mapping network including the spatial distribution features and the device posture parameters, and calculating a spatial error vector between the observed posture of the current image area and the expected posture of the device based on the spatial relationship mapping network, comprises: Establishing a multidimensional mapping topology structure with the coordinates of the center point of the image area and the coordinates of the device end posture as nodes, and using the node connection weights in the multidimensional mapping topology structure as the matching degree between the geometric parameters of the area and the tool orientation angle; Constructing an error transmission link between nodes based on the multidimensional mapping topology structure, and transmitting the relative distance distribution features in the spatial distribution features to the terminal pose coordinate nodes in sequence through the error transmission link; Introducing the environmental interference feature as a link correction factor to each intermediate node in the error conduction link, and updating the distribution value of the node connection weight; The spatial error vector is back-propagated according to the updated node connection weights to obtain an initial spatial error vector between the observed pose of each image region and the expected pose of the device.

6. The processing positioning method of adaptive image recognition according to claim 4, characterized in that: The performing interference weight allocation processing on the environmental interference feature, and correcting the spatial error vector according to the interference weight allocation processing result to generate a corrected spatial error correction amount, includes: Determining the illumination intensity gradient distribution of the image region according to the imaging illumination distribution uniformity characteristic, and calculating the influence factor of the illumination intensity gradient distribution on the positioning accuracy of the target object contour; Extracting the target noise amplitude and target interference distribution pattern from the background interference noise characteristics, and determining the influence weight of the target noise on the edge positioning accuracy and the influence weight of the target interference on the stability of the region segmentation; Analyzing the blur direction and blur length in the blur feature, and combining the actual motion speed in the device motion trajectory data, calculating the error contribution weight of the blur to the image analysis feature; The positioning accuracy influencing factor, the influence weight of the target noise on the edge positioning accuracy, the influence weight of the target interference on the regional segmentation stability, and the error contribution weight are standardized to generate a comprehensive interference weight distribution matrix; A weighted correction operator is generated based on the comprehensive interference weight distribution matrix, and the weighted correction operator is applied to the initial spatial error vector to obtain the corrected spatial error correction amount.

7. The processing positioning method of adaptive image recognition according to claim 6, characterized in that: The generating the spatial positioning drive feature set by combining the corrected spatial error correction amount includes: Obtaining positioning accuracy constraints in the target machining task parameters, wherein the positioning accuracy constraints include an upper tolerance limit for axial positioning error, an angle deviation threshold, and an allowable fluctuation range for contact pressure; Calculating a proportionality coefficient between the corrected spatial error correction amount and the upper tolerance limit of the axial positioning error, and determining an adaptive adjustment step length of the displacement compensation amount of the end of the manipulator according to the proportionality coefficient; Performing dead-zone filtering on the tool posture rotation angle according to the angle deviation threshold, setting the rotation angle component smaller than the angle deviation threshold to zero, and filtering out slight angle jitter interference; Applying a boundary constraint of the allowable fluctuation range of the contact pressure to the adjustment coefficient of the processing pressure, and clipping the adjustment coefficient to the nearest boundary value when it exceeds the boundary constraint; A spatial positioning drive feature set that meets the positioning accuracy constraint condition is generated according to the adaptive adjustment step size, the tool posture rotation angle after the dead zone filtering process, and the trimmed adjustment coefficient.

8. The processing positioning method of adaptive image recognition according to claim 4, characterized in that: The step of generating an adaptive positioning control instruction according to the spatial positioning drive feature set and transmitting the instruction to the processing equipment execution end to instruct the processing equipment to perform an adaptive posture positioning control operation for the current image area includes: Generating a path planning intermediate node sequence according to the displacement compensation amount of the end of the robot arm, and inserting a smooth transition trajectory into the path planning intermediate node sequence to eliminate the sudden stop jitter during the movement of the robot arm; Converting the tool posture rotation angle into a quaternion representation and associating it with each node in the path planning intermediate node sequence to generate a continuous posture control instruction stream for the end effector of the robot arm; performing time-domain discretization processing on the adjustment coefficient of the processing pressure to obtain a pressure adjustment gradient curve for each time segment, and coupling the pressure adjustment gradient curve with a servo motor control signal of the processing tool; Combining the continuous posture control instruction stream and the coupled servo motor control signal into a unified time synchronization control instruction set, and encapsulating it into an adaptive positioning control instruction data packet according to the processing equipment communication protocol; Transmitting the adaptive positioning control instruction data packet to an execution end of a processing device via a real-time bus, instructing the execution end to parse the data packet and drive the robotic arm and the processing tool to perform a collaborative positioning operation; After the instruction processing device performs the adaptive posture positioning control operation on the current image area, the method further includes: Real-time capture of actual posture feedback data from the execution end of the processing equipment and quality inspection image data of the processing surface; performing a deviation analysis between the actual posture feedback data and the expected parameters in the spatial positioning drive feature set, and triggering reacquisition of the multi-source image data set to perform a new round of adaptive positioning control when the deviation exceeds an adjustment threshold; Performing surface defect recognition processing on the quality inspection image data; when defects exceeding the allowable standard are detected on the processed surface, backtracking the corresponding spatial positioning drive feature set based on the defect distribution characteristics to perform parameter tuning learning and update the positioning control strategy in subsequent processing tasks; The actual posture feedback data is associated with the quality inspection image data and stored in a processing knowledge base as an optimized training sample for generating adaptive positioning control instructions in the next batch of processing tasks.

9. A processing positioning system, characterized in that: include: Memory, used to store program instructions and data; A processor, coupled to a memory, and configured to execute instructions in the memory to implement the method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that The method comprises instructions which, when executed on a processor, implement the method according to any one of claims 1 to 8.

Images