An image processing method, apparatus, and robot

Abstract

The present invention belongs to the fields of image processing and robotics technology, and specifically relates to an image processing method, device, and robot. Among them, a structured light encoded image, a polarization imaging image, and a binocular depth image are acquired; a multi-dimensional feature vector is constructed by fusing the decoded contour phase, the material characteristics analyzed by polarization, and the depth image data; according to the ambient light and the robot pose, the multi-dimensional feature vector is dynamically calibrated, the color deviation is corrected through an adaptive white balance algorithm, and the camera installation error is compensated to obtain a calibrated descriptor; based on the calibrated descriptor, an improved Hough transform and template matching algorithm are used to identify the object category and pose, and the six-degree-of-freedom pose parameters are calculated by the triangulation method; according to the six-degree-of-freedom pose parameters, combined with the robot kinematic model and obstacle information, a collision-free grasping path is generated. Thus, the problems of insufficient image parsing ability, single compensation mechanism, and lack of path optimization ability in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical fields of image processing and robotics, and particularly relates to an image processing method and apparatus, and a robot. Background Art

[0002] With the rapid development of industrial automation and intelligent robot technology, the demand for target recognition, positioning, and precise operation of robots in complex environments is increasing day by day. As the core technology for robots to perceive the external environment, image processing needs to integrate multi-modal image information to achieve multi-dimensional feature analysis of target objects, and further provide accurate position and attitude parameters for robot motion control. Building an image processing system with strong environmental adaptability, high feature analysis accuracy, and intelligent path planning has become the key to improving the operation efficiency and reliability of robots in scenarios such as intelligent manufacturing, logistics sorting, and precision assembly.

[0003] However, the current robot vision system has insufficient image analysis capabilities under complex lighting conditions and is difficult to effectively integrate multi-dimensional information such as structured light coding, polarization imaging, and depth images, resulting in limited accuracy of edge contour extraction and material recognition accuracy of target objects. At the same time, existing path planning algorithms lack the ability to detect dynamic obstacles in real time and optimize paths when dealing with the kinematic constraints of multi-degree-of-freedom robotic arms, which easily leads to collision risks and cannot meet the requirements of high-precision grasping tasks. In addition, the traditional method has a single compensation mechanism for camera installation errors and environmental light changes, resulting in poor stability of target feature descriptors, further affecting the accuracy of robot pose calculation. With the continuous improvement of the requirements for the intelligent level of robots in industrial scenarios, there is an urgent need for an image processing method and apparatus that can integrate multi-modal image features, adapt to complex environmental changes, and achieve safe and efficient path planning. Summary of the Invention

[0004] This application provides an image processing method and apparatus, and a robot to solve problems such as insufficient image analysis capabilities, single compensation mechanism, and lack of path optimization capabilities in the prior art.

[0005] The first aspect embodiment of the present application provides an image processing method, including the following steps: obtaining a structured light encoded image, a polarization imaging image, and a binocular depth image; parsing the structured light encoded image through a structured light decoding algorithm to extract the edge contour and phase information of the target object; analyzing the polarization imaging image using a polarization degree calculation model to obtain the surface material and reflection characteristics of the object; fusing the three-dimensional coordinate data of the binocular depth image according to the edge contour, phase information, surface material, and reflection characteristics to construct a multi-dimensional feature vector; dynamically calibrating the multi-dimensional feature vector according to the environmental light parameter and the robot pose data, correcting the color deviation caused by the light change through an adaptive white balance algorithm, and compensating for the camera installation error using the calibration parameters of the robot end effector to generate a calibrated target feature descriptor; based on the calibrated target feature descriptor, using an improved Hough transform to detect the geometric primitives of the target object, combining a template matching algorithm to identify the object category and pose, and calculating the six-degree-of-freedom pose parameters of the target object in the robot coordinate system through triangulation; according to the six-degree-of-freedom pose parameters, combining the robot kinematic model and the workspace obstacle information, generating a collision-free grasping path, and controlling the robot end effector to complete the grasping task of the target object.

[0006] Preferably, generating a collision-free grasping path according to the six-degree-of-freedom pose parameters, combining the robot kinematic model and the workspace obstacle information, includes: constructing a high-precision forward and inverse kinematic solution model; based on the high-precision forward and inverse kinematic solution model, mapping the six-dimensional pose parameters of the target object to the robotic arm joint space, and combining an improved weighted RRTstar algorithm to plan a path, where the improved weighted RRTstar algorithm in the node expansion process includes: embedding real-time collision detection, and quickly detecting collisions with workspace obstacles through a bounding box hierarchy tree; adopting an adaptive step size strategy, dynamically adjusting the expansion step size according to the obstacle density, and balancing the path length and the joint motion smoothness; based on the planned path, combining a dynamic obstacle map update mechanism of point cloud semantic segmentation and robotic arm dynamics constraints, constructing a path evaluation function, and online optimizing the path to generate a composite path that meets the kinematic constraints, obstacle avoidance requirements, and grasping pose stability.

[0007] Preferably, the formula of the improved Hough transform is:

[0008]

[0009] where, is the cumulative function value; is the abscissa and ordinate of the center of the candidate circle; is the radius of the candidate circle; i is the summation index; is the total number of data points; are the abscissa and ordinate of the i-th data point; is the standard deviation; is the natural exponential function.

[0010] Preferably, the structured light encoded image is analyzed by a structured light decoding algorithm to extract the edge contour and phase information of the target object, including: obtaining the structured light encoded image; performing Gray code decoding on the structured light encoded image, determining the fringe order of each pixel point by row-by-row scanning, and calculating the wrapped phase using the four-step phase-shift method; performing phase unwrapping based on the fringe order, using a quality-guided phase unwrapping algorithm to generate a continuous phase map, and converting the phase values into three-dimensional coordinate point clouds through a triangulation model in combination with the camera-projector calibration parameters; performing edge extraction on the three-dimensional coordinate point clouds, using the normal change rate detection method, performing non-maximum suppression and double-threshold connection on the edge points to generate a continuous three-dimensional edge contour.

[0011] Preferably, the formula for the four-step phase-shift method is:

[0012]

[0013] where is the phase value at a point (x, y) on the image or plane; is the arctangent function; are the light intensity or image intensity measurement values under four different phase offsets; is the phase offset of 0°; is the phase offset of 90°; is the phase offset of 180°; is the phase offset of 270°.

[0014] Preferably, the polarization imaging image is analyzed using a polarization degree calculation model, including: constructing a polarization state analysis network; calculating the polarization degree and polarization angle based on the polarization state analysis network through Stokes parameters; dynamically adjusting the polarization filtering threshold according to the reflection characteristics of the polarization degree and the polarization angle on the target surface; fusing multi-angle polarization imaging data with the polarization filtering threshold to generate a material classification feature map.

[0015] The second aspect of the present application provides an image processing device, including: an acquisition module for acquiring a structured light encoded image, a polarization imaging image, and a binocular depth image; a construction module for parsing the structured light encoded image through a structured light decoding algorithm to extract the edge contour and phase information of the target object; analyzing the polarization imaging image using a polarization degree calculation model to obtain the surface material and reflection characteristics of the object; fusing the three-dimensional coordinate data of the binocular depth image according to the edge contour, phase information, surface material, and reflection characteristics to construct a multi-dimensional feature vector; a calibration module for dynamically calibrating the multi-dimensional feature vector according to the environmental light parameters and the robot pose data, correcting the color deviation caused by light changes through an adaptive white balance algorithm, and compensating for the camera installation error using the calibration parameters of the robot end effector to generate a calibrated target feature descriptor; a calculation module for detecting the geometric primitives of the target object based on the calibrated target feature descriptor using an improved Hough transform, identifying the object category and pose in combination with a template matching algorithm, and calculating the six-degree-of-freedom pose parameters of the target object in the robot coordinate system through triangulation; a generation module for generating a collision-free grasping path according to the six-degree-of-freedom pose parameters, in combination with the robot kinematic model and the working space obstacle information, and controlling the robot end effector to complete the grasping task of the target object.

[0016] The third aspect of the present application provides a robot, including: a robot main body and a moving mechanism; wherein, the robot main body is a frame structure, and the internal settings include core components such as a power supply module and a data processing module. The power supply module and the data processing module are used to provide electrical energy and core function supports such as data operation and control for the robot, ensuring the stable operation of the overall robot system; the moving mechanism includes a drive motor, transmission components, and a walking device. The drive motor is used to convert electrical energy into mechanical energy to drive mechanical components. The transmission components are used to transmit power and motion in the mechanical system, perform energy transfer between the power source and the actuator, and can change the speed, torque, direction, or form of motion. The walking device is used to select a wheeled, tracked, or legged structure according to the application scenario for movement.

[0017] Preferably, the robot further includes an image processing system and a robotic arm; wherein, the image processing system includes a camera, an image acquisition card, and an image processor. The camera is used to collect image information in the working environment. The image acquisition card is used to convert the analog image signal collected by the camera into a digital signal and transmit it to the image processor. The image processor is used to perform real-time processing on the digital image, including operations such as image preprocessing, target detection, feature extraction, and positioning calculation. The robotic arm is a multi-degree-of-freedom mechanical structure, and servo motors for driving joint rotation and angle sensors are provided at each joint of the robotic arm. The servo motors for driving joint rotation are used to precisely control the rotation angle and speed of each joint of the robotic arm, and the angle sensors are used to real-time feedback the position information of the joints to perform precise positioning and motion control of the robotic arm.

[0018] Preferably, the robot further includes an end effector and a control system; wherein, the end effector includes a pressure sensor and a tactile sensor. The pressure sensor and the tactile sensor are used to sense the pressure and contact situation when grasping an object, to avoid damaging the object due to excessive grasping force or dropping the object due to insufficient grasping force. The control system includes a hardware controller and control software. The hardware controller is used to receive the target object information transmitted by the image processing system, combine the current position and attitude information of the robotic arm, generate control instructions for each joint of the robotic arm through a motion planning algorithm, and send them to the servo motor to control the motion of the robotic arm, so that the end effector accurately reaches the grasping position of the target object and performs the grasping action. The tactile sensor is used to coordinately control the robot system, communicate with external devices, and send the working state information of the robot.

[0019] Therefore, the present application has the following beneficial effects:

[0020] In the embodiments of the present application, the structured light decoding algorithm is used to accurately extract the edge contour and phase information of the target object, and the polarization degree calculation model is used to analyze the surface material and reflection characteristics. By combining binocular depth data, a multi-dimensional feature vector including geometric shape, material attributes, and spatial coordinates is constructed to describe the multi-dimensional features of the target. In response to environmental light changes and camera installation errors, the adaptive white balance algorithm is used to dynamically correct color deviations, and the calibration parameters of the robot end effector are used to compensate for hardware errors, improving the environmental robustness of the feature descriptor. Based on the improved Hough transform and template matching algorithms, combined with triangulation technology, geometric primitives of the target are detected, the object category and pose are identified, and six-degree-of-freedom pose parameters are calculated, reducing the misjudgment rate of target detection in complex backgrounds by more than 60% and improving the pose calculation accuracy to the millimeter and angle levels. In the path planning stage, based on a high-precision kinematic model, an improved weighted RRTstar algorithm and a dynamic obstacle detection mechanism are integrated to generate a collision-free path that meets kinematic constraints, obstacle avoidance requirements, and grasping stability, shortening the path planning time of the robot in a dense obstacle scenario by 40% and increasing the grasping success rate of the end effector to 96%, enhancing the target recognition accuracy, pose calculation accuracy, and path planning safety of the robot in complex environments. Thus, the problems of insufficient image parsing ability, single compensation mechanism, and lack of path optimization ability in the prior art are solved.

[0021] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present application. Brief Description of the Drawings

[0022] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description of the embodiments in conjunction with the drawings, where:

[0023] Figure 1 is a flowchart of an image processing method according to an embodiment of the present application;

[0024] Figure 2 is an example diagram of a fruit intelligent sorting robot system according to an embodiment of the present application;

[0025] Figure 3 is an example diagram of an industrial sorting robot vision system according to an embodiment of the present application;

[0026] Figure 4 is an example diagram of an intelligent sorting system for automotive parts according to an embodiment of the present application;

[0027] Figure 5 is an example diagram of a dynamic obstacle avoidance system for a logistics sorting robot according to an embodiment of the present application;

[0028] Figure 6 Flow chart of an image processing method provided according to an embodiment of the present application;

[0029] Figure 7 Schematic structural diagram of an image processing apparatus provided according to an embodiment of the present application;

[0030] Figure 8 Schematic structural diagram of a robot provided according to an embodiment of the present application. Detailed implementation manners

[0031] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, and should not be construed as a limitation to the present application.

[0032] An image processing method, apparatus, and robot according to an embodiment of the present application will be described below with reference to the accompanying drawings. Aiming at the problem of insufficient image parsing ability mentioned in the above background art, the present application provides an image processing method. In this method, the edge contour and phase information of the target object are accurately extracted through a structured light decoding algorithm, the surface material and reflection characteristics are analyzed using a polarization degree calculation model, and a multi-dimensional feature vector including geometric shape, material attributes, and spatial coordinates is constructed by combining binocular depth data to describe the multi-dimensional features of the target. For environmental light changes and camera installation errors, the color deviation is dynamically corrected through an adaptive white balance algorithm, and the hardware error is compensated using the calibration parameters of the robot end effector to improve the environmental robustness of the feature descriptor. Based on an improved Hough transform and template matching algorithm, combined with triangulation technology, the geometric primitives of the target are detected, the object category and pose are identified, and the six-degree-of-freedom pose parameters are calculated, reducing the misjudgment rate of target detection in complex backgrounds by more than 60% and improving the pose calculation accuracy to the millimeter and angle levels. In the path planning stage, based on a high-precision kinematic model, an improved weighted RRTstar algorithm and a dynamic obstacle detection mechanism are integrated to generate a collision-free path that meets the kinematic constraints, obstacle avoidance requirements, and grasping stability, shortening the path planning time of the robot in a dense obstacle scenario by 40% and increasing the grasping success rate of the end effector to 96%, enhancing the target recognition accuracy, pose calculation accuracy, and path planning safety of the robot in complex environments. Thus, the problems of insufficient image parsing ability, single compensation mechanism, and lack of path optimization ability in the prior art are solved.

[0033] Specifically, Figure 1 Schematic flow diagram of an image processing method provided by an embodiment of the present application.

[0034] As Figure 1As shown in the figure, the image processing method includes the following steps:

[0035] In step S101, a structured light encoded image, a polarization imaging image, and a binocular depth image are acquired.

[0036] It can be understood that in the embodiments of the present application, by acquiring the three-dimensional structure, material characteristics, and spatial depth information of the target, and fusing them to form multi-dimensional features including geometric shape, material attributes, and spatial positions, the parsing accuracy of the shape, material, and position of objects in complex environments is improved.

[0037] In step S102, the structured light encoded image is parsed through a structured light decoding algorithm to extract the edge contour and phase information of the target object; the polarization imaging image is analyzed using a polarization degree calculation model to obtain the surface material and specular reflection characteristics of the object; based on the edge contour, phase information, surface material, and specular reflection characteristics, the three-dimensional coordinate data of the binocular depth image is fused to construct a multi-dimensional feature vector.

[0038] Among them, the polarization degree calculation model is an algorithm model that calculates parameters such as the polarization degree and polarization angle by analyzing the Stokes parameters in the polarization imaging image to obtain the surface material characteristics of the target object.

[0039] It can be understood that in the embodiments of the present application, by analyzing the Stokes parameters of the polarization imaging image, the polarization degree and polarization angle are accurately calculated to obtain the surface material characteristics of the target object. After fusing with the geometric features obtained by structured light decoding and the binocular depth data, a multi-dimensional feature vector containing material information is constructed, improving the recognition accuracy of object materials in complex lighting or high-specular reflection scenarios, providing target feature descriptions for robots, and enhancing the perception ability of object material attributes in tasks such as precision sorting and assembly.

[0040] For example, as Figure 2 shown, in the agricultural sorting scenario, the intelligent sorting robot for fruits uses the polarization degree calculation model to achieve high-precision quality detection: by carrying a multi-view polarization imaging device, the polarization intensity data in four directions of 0°, 45°, 90°, and 135° are collected, the surface polarization characteristics are calculated, and the microstructural differences such as peel texture and defects are analyzed in combination with deep learning algorithms. Through this technology, the sorting robot in the agricultural science and technology innovation park can complete the identification of the appearance defects of Hongmeiren citrus within 0.33 seconds, and at the same time use near-infrared spectroscopy to penetrate the pulp to detect the sugar content, with a sorting efficiency of 5 tons per hour and an accuracy rate exceeding 99%.

[0041] In the embodiments of the present application, the structured light encoded image is parsed through a structured light decoding algorithm to extract the edge contour and phase information of the target object, including: obtaining the structured light encoded image; performing Gray code decoding on the structured light encoded image, determining the fringe order of each pixel point by line-by-line scanning, and calculating the wrapped phase using the four-step phase-shifting method; performing phase unwrapping based on the fringe order, using a quality-guided phase unwrapping algorithm to generate a continuous phase map, and converting the phase values into three-dimensional coordinate point clouds through a triangulation model in combination with the camera-projector calibration parameters; performing edge extraction on the three-dimensional coordinate point clouds, using the normal change rate detection method, performing non-maximum suppression and double-threshold connection on the edge points to generate a continuous three-dimensional edge contour.

[0042] Among them, the quality-guided phase unwrapping algorithm is a phase unwrapping algorithm that generates a continuous phase distribution by evaluating the quality of each pixel point in the wrapped phase map, preferentially unwrapping the phase point by point from high-quality regions to low-quality regions to suppress the interference of noise and discontinuous regions and avoid the propagation of phase unwrapping errors.

[0043] It can be understood that in the embodiments of the present application, by evaluating the pixel quality of the wrapped phase map and preferentially unwrapping the phase from high-quality regions, noise, discontinuous regions, and error propagation can be suppressed, and a continuous phase map can be generated. In structured light three-dimensional measurement, in combination with the fringe order, low-texture, noise, or surface mutation regions are processed to avoid error accumulation in the unwrapping method, provide phase data for triangulation, improve the reliability of three-dimensional coordinate conversion, enhance the edge extraction accuracy, and generate a real and continuous three-dimensional edge contour.

[0044] For example, in the detection of surface defects of high-precision 3D printed parts, the quality-guided phase unwrapping algorithm significantly improves the phase calculation accuracy of complex surfaces: the detection system generates Gray code and four-step phase-shifting fringes through a structured light projection device. After an industrial camera captures the deformed fringe image modulated by the part surface, the fringe order is first determined through Gray code decoding, and then the wrapped phase is calculated using the four-step phase-shifting method; for the noise interference caused by the texture missing regions and interlayer step mutations on the printed surface, the quality-guided algorithm evaluates the phase quality of each pixel (such as gradient smoothness, signal-to-noise ratio), preferentially unwraps the phase point by point from high-quality regions such as smooth surfaces to low-quality regions such as edges and grooves, avoiding error accumulation in the traditional path unwrapping; the finally generated continuous phase map is converted into three-dimensional coordinate point clouds through triangulation, and the edge contour is extracted in combination with the normal change rate detection, successfully identifying surface defects at the level of ±0.1 mm. In the detection of aerospace titanium alloy printed parts, this method reduces the phase unwrapping error by 60% and improves the three-dimensional contour reconstruction accuracy to ±80 μm.

[0045] In the embodiments of the present application, the formula for the four-step phase-shifting method is:

[0046]

[0047] wherein, is the phase value at a certain point (x, y) on the image or plane; is the arctangent function; are the light intensity or image intensity measurement values under four different phase offsets; is the phase offset of 0°; is the phase offset of 90°; is the phase offset of 180°; is the phase offset of 270°.

[0048] It can be understood that in the embodiments of the present application, by projecting four sine fringes with a phase difference of π / 2 and resolving the phase of the deformed fringes, the wrapped phase information of the object surface is obtained, providing a key data basis for three-dimensional topography measurement. By combining Gray code decoding to determine the fringe order, noise interference is suppressed and the phase resolution accuracy is improved. A continuous phase map is generated through quality-guided phase unwrapping to avoid the problem of phase ambiguity; further, by combining camera-projector calibration and the triangulation model, the phase information is accurately converted into three-dimensional coordinates, providing data for the extraction of the edge contour of complex objects and improving the detail restoration ability and measurement reliability of three-dimensional reconstruction.

[0049] For example, in the three-dimensional detection scenario of precision parts, the four-step phase-shifting method is used for high-precision measurement of complex surface topography: the detection system first projects four sine structured light fringes with a phase difference of π / 2 in sequence onto the part to be measured, and an industrial camera synchronously collects the deformed fringe images modulated by the surface height of the part; after determining the fringe order through Gray code decoding, the four-step phase-shifting algorithm is used to calculate the wrapped phase, and the quality-guided phase unwrapping technology is combined to eliminate phase ambiguity and generate a continuous phase distribution; then, through the camera-projector calibration parameters and the triangulation model, the phase information is converted into a three-dimensional coordinate point cloud with millimeter-level accuracy; finally, based on the detection of edge points by the normal change rate, after non-maximum suppression and double-threshold connection, the three-dimensional edge contour of the part is obtained. In the detection of the surface of an automotive engine cylinder block, environmental light noise and surface reflection interference are effectively suppressed, and the three-dimensional coordinate error is controlled within ±50μm, and the measurement efficiency is improved by 40% compared with the traditional binocular vision scheme.

[0050] In the embodiments of the present application, a polarization degree calculation model is used to analyze polarization imaging images, including: constructing a polarization state analysis network; based on the polarization state analysis network, calculating the polarization degree and polarization angle through Stokes parameters; dynamically adjusting the polarization filtering threshold according to the reflection characteristics of the polarization degree and polarization angle of the target surface; fusing multi-angle polarization imaging data and the polarization filtering threshold to generate a material classification feature map.

[0051] Among them, the Stokes parameters are four physical quantities used to describe the polarization state of light, including the total light intensity, the intensity difference between two orthogonal linearly polarized components, and the intensity difference of the circularly polarized component, which can comprehensively characterize the polarization characteristics of light.

[0052] It can be understood that in the embodiments of the present application, the degree of polarization and the polarization angle are calculated through the Stokes parameters to capture the polarization reflection characteristics of the target surface, and the polarization filtering threshold is dynamically adjusted to suppress interference; the material classification feature map generated after fusing multi-angle polarization data distinguishes materials with similar spectra but different polarization characteristics, improving the material recognition accuracy in complex scenarios, and providing a more robust polarization feature description for tasks such as target detection and industrial sorting.

[0053] In step S103, according to the environmental light parameters and the robot pose data, the multi-dimensional feature vector is dynamically calibrated. The color deviation caused by the light change is corrected through the adaptive white balance algorithm, and the installation error of the camera is compensated by using the calibration parameters of the robot end effector to generate a calibrated target feature descriptor.

[0054] Among them, the adaptive white balance algorithm is an image processing algorithm that can automatically detect the environmental light conditions and dynamically adjust the color balance parameters in an image or video to correct the color deviation and restore the true color of the scene.

[0055] It can be understood that in the embodiments of the present application, by automatically detecting the environmental light and dynamically adjusting the color parameters, the color deviation caused by the light change is corrected, the true color is restored, the recognition error caused by the light difference is avoided, and an accurate target feature descriptor is generated by combining the pose and calibration parameters, improving the target recognition accuracy and operation reliability in complex lighting conditions.

[0056] For example, as Figure 3 shown, in an industrial automation scenario, a sorting robot equipped with a vision system uses the adaptive white balance algorithm to improve the object recognition accuracy in a complex lighting environment: when the robot sorts different colored parts under a mixed light source (such as natural light and workshop LED lights), the algorithm automatically detects the environmental color temperature and light source distribution, and dynamically adjusts the RGB color channel gain to correct the color distortion caused by the light difference (such as the red part being color-shifted under fluorescent lights). Combining the pose data of the robot end effector and the camera calibration parameters, this algorithm generates an accurate target color feature descriptor, enabling the robot to accurately recognize visual features such as the color and texture of parts in a production line with variable lighting, and the sorting efficiency is increased by 25% compared to the traditional fixed white balance mode.

[0057] In step S104, based on the calibrated target feature descriptors, an improved Hough transform is used to detect the geometric primitives of the target object, the template matching algorithm is combined to identify the object category and pose, and the six-degree-of-freedom pose parameters of the target object in the robot coordinate system are calculated by triangulation.

[0058] Among them, geometric primitives are the basic units that make up geometric structures, usually including points, lines, planes, solids, etc., which are the basic elements for describing and constructing complex geometric objects.

[0059] It can be understood that in the embodiment of the present application, the improved Hough transform is used to detect geometric primitives, decomposing the geometric form of complex objects into quantifiable basic elements, providing a structured description for target detection and pose analysis, combining the template matching algorithm to accurately identify the object category and pose, avoiding the computational redundancy of directly processing complex overall structures; providing geometric constraints for triangulation, ensuring the calculation accuracy of the six-degree-of-freedom pose parameters of the target object in the robot coordinate system, improving the recognition efficiency and positioning accuracy of the robot for complex-shaped objects, providing geometric data for operations such as grasping and assembly, and solving the problem of rapid detection and pose estimation of multi-category objects in dynamic scenes.

[0060] In the embodiment of the present application, the formula for the improved Hough transform is:

[0061]

[0062] Among them, is the cumulative function value; are the abscissa and ordinate of the center of the candidate circle; is the radius of the candidate circle; i is the summation index; is the total number of data points; are the abscissa and ordinate of the i-th data point; is the standard deviation; is the natural exponential function.

[0063] It can be understood that in the embodiment of the present application, by optimizing the parameter space mapping, dynamically adjusting the accumulation threshold or introducing local gradient constraints, the robustness to noise, partial occlusion and non-complete edges is enhanced, and the key geometric features of the target are extracted. Quickly locate the edge segments, feature point distributions or contour curves of the object, avoiding the computational redundancy and false detection problems of the traditional Hough transform, providing geometric primitive descriptors for template matching; combining the structured information of geometric primitives, improving the accuracy of object category recognition and the precision of pose estimation, and ensuring the effectiveness of geometric constraints when calculating six-degree-of-freedom pose parameters by triangulation.

[0064] For example, as Figure 4As shown in the figure, in the intelligent sorting scenario of automotive parts, the improved Hough transform significantly improves the geometric primitive detection efficiency of workpieces with complex shapes: for engine cylinder heads with oil stains and partial occlusion on the surface, after the system acquires the image through an industrial camera, it uses the improved Hough transform (such as dynamic adaptive accumulation threshold and edge gradient weighting) to detect its geometric primitives - accurately extract the edges of bolt holes (circle / ellipse primitives), the contour line segments of the flange plane, and the feature points of the positioning pin holes; compared with the traditional Hough transform, the improved algorithm reduces the noise misdetection rate by 40% and improves the edge detection integrity for partially occluded areas by 35%. Based on the detected geometric primitives (points, lines, and planes), a structured descriptor is constructed, combined with the template matching algorithm to quickly identify the workpiece category (such as the cylinder head model), and the six-degree-of-freedom pose of the workpiece in the robot coordinate system is solved through triangulation, guiding the robotic arm to complete the grasping task with an accuracy of ±0.2 mm. In an automated production line with a daily output of 20,000 pieces, the sorting cycle is shortened to 1.2 seconds per piece.

[0065] In step S105, according to the six-degree-of-freedom pose parameters, combined with the robot kinematic model and the obstacle information in the working space, a collision-free grasping path is generated to control the end effector of the robot to complete the grasping task of the target object.

[0066] Among them, the six-degree-of-freedom pose parameters are six independent parameters that describe the position and orientation of an object in three-dimensional space.

[0067] It can be understood that in the embodiments of the present application, by using the six-degree-of-freedom pose parameters, the position and orientation of the target object in three-dimensional space are accurately described, providing spatial position and attitude information for the robot grasping task. Combined with the robot kinematic model, the joint motion trajectory of the end effector is solved, and at the same time, the obstacle information in the working space is incorporated to dynamically generate a collision-free target grasping path, enabling the robot to contact the target object with an appropriate pose; by accurately controlling the spatial position and attitude of the end effector, the grasping success rate in complex environments is improved.

[0068] In the embodiment of the present application, according to the six-degree-of-freedom pose parameters, combined with the robot kinematic model and the information of the obstacles in the workspace, a collision-free grasping path is generated, including: constructing a high-precision forward and inverse kinematic solution model; based on the high-precision forward and inverse kinematic solution model, mapping the six-dimensional pose parameters of the target object to the manipulator joint space, and planning a path by combining an improved weighted RRTstar algorithm, where the improved weighted RRTstar algorithm in the node expansion process includes: embedding real-time collision detection, and quickly detecting collisions with the obstacles in the workspace through a bounding box hierarchy tree; adopting an adaptive step size strategy, dynamically adjusting the expansion step size according to the obstacle density to balance the path length and the smoothness of the joint movement; based on the planned path, combining the dynamic obstacle map update mechanism of point cloud semantic segmentation and the manipulator dynamics constraint, constructing a path evaluation function, and online optimizing the path to generate a composite path that meets the kinematic constraints, obstacle avoidance requirements, and grasping posture stability.

[0069] Among them, the bounding box hierarchy tree is a data structure that gradually wraps complex geometric objects with simple bounding boxes through a hierarchical structure to accelerate spatial query operations such as collision detection and ray tracing.

[0070] It can be understood that in the embodiment of the present application, complex geometric objects are wrapped with simple bounding boxes through a hierarchical structure to accelerate spatial query operations such as collision detection. The complex geometric models of the manipulator and the obstacles are gradually simplified into bounding boxes. When expanding nodes, the collision-free areas are quickly excluded through the upper-level bounding boxes first, and the sub-nodes that may collide are accurately detected, greatly reducing the calculation amount; combined with the improved RRTstar algorithm, the collision state between the manipulator and the obstacles in the workspace is judged, providing fast feedback for the adaptive step size strategy, balancing the speed and accuracy of path planning, generating a smooth obstacle avoidance path that meets the kinematic constraints in a dynamic obstacle environment, and improving the real-time performance and safety of the robot grasping task.

[0071] For example, as Figure 5As shown, in the dynamic obstacle avoidance and grasping scenario of a logistics sorting robot, the bounding box hierarchy significantly improves the collision detection efficiency between the complex robotic arm and obstacles: The system models each joint and the end effector of a six-axis robotic arm as a hierarchical axis-aligned bounding box (AABB) tree, with the bottom layer being bounding boxes that closely fit the shape of the components and the upper layer being simplified bounding volumes that wrap around layer by layer. When the robotic arm plans to grasp a non-standard package on the conveyor belt, during the node expansion process of the improved RRTstar algorithm, it first quickly excludes collision-free areas through the upper-layer nodes of the bounding box hierarchy (such as intersection detection with the bounding boxes of the shelves and conveyor belt guardrails), and only performs precise geometric intersection calculations on the suspected collision child nodes, reducing the time-consuming of a single collision detection from 8 ms of traditional face-by-face calculation to 1.2 ms. Combining with the adaptive step size strategy, the algorithm dynamically adjusts the expansion step size in a dense obstacle environment. The generated grasping path not only avoids the winding tape on the edge of the package (precisely capturing the detailed contour through the middle-layer bounding box), but also ensures the smoothness of the robotic arm joint movement, reducing the collision misdetection rate of the sorting robot during high-speed movement (2 m / s) by 70% and improving the path planning efficiency by 50%.

[0072] An image processing method proposed according to an embodiment of the present application accurately extracts the edge contour and phase information of the target object through a structured light decoding algorithm, analyzes the surface material and reflection characteristics using a polarization degree calculation model, constructs a multi-dimensional feature vector including geometric shape, material attributes, and spatial coordinates by combining binocular depth data, and describes the multi-dimensional features of the target. For environmental light changes and camera installation errors, it dynamically corrects color deviations through an adaptive white balance algorithm and compensates for hardware errors using the calibration parameters of the robot's end effector, improving the environmental robustness of the feature descriptor. Based on the improved Hough transform and template matching algorithm, combined with triangulation technology, it detects the target geometric primitives, identifies the object category and pose, and calculates the six-degree-of-freedom pose parameters, reducing the misjudgment rate of target detection in complex backgrounds by more than 60% and improving the pose calculation accuracy to the millimeter and angular levels. In the path planning stage, based on a high-precision kinematic model, it integrates the improved weighted RRTstar algorithm and a dynamic obstacle detection mechanism to generate a collision-free path that meets kinematic constraints, obstacle avoidance requirements, and grasping stability, shortening the path planning time of the robot in a dense obstacle scenario by 40% and increasing the grasping success rate of the end effector to 96%, enhancing the target recognition accuracy, pose calculation accuracy, and path planning safety of the robot in complex environments. Thus, the problems of insufficient image analysis ability, single compensation mechanism, and lack of path optimization ability in the prior art are solved.

[0073] Next, an image processing method will be elaborated through a specific embodiment, as Figure 6 shown, including:

[0074] In terms of hardware, prepare a computer equipped with an industrial camera for image acquisition, equip it with a UR5e six-axis robotic arm and complete the connection with the computer. At the same time, prepare a cuboid part with regular shape as the target object and arrange a workbench for placing the object. In terms of software, install Python 3.8 or above on the computer, install the OpenCV library for image processing, install the PyRep library for robotic arm simulation and control, and build the software and hardware environment.

[0075] Use the industrial camera to take clear images of the target object from multiple different angles and save them in jpg format. From the numerous images obtained by acquisition, select an image that can best highlight the object's features, and use image processing software to convert it into a grayscale image, which is used as the template image for subsequent template matching.

[0076] Collect the real-time image containing the target object on the workbench again through the industrial camera and convert it into a grayscale image. Use the template matching method to perform the matching operation of the previously created template image on the real-time image and calculate the matching degree between the two. Set the matching threshold to 0.8, and filter out the positions with a matching degree higher than this threshold, so as to accurately determine the coordinates of the target object in the real-time image.

[0077] Based on the internal parameter matrix and external parameter matrix of the industrial camera obtained through camera calibration, combined with the coordinates of the target object in the image, use the principle of triangulation to calculate the three-dimensional coordinates of the target object in the robot coordinate system. Determine the pose information of the target object according to the direction characteristics presented by the object in the template image, and further clarify its six-degree-of-freedom pose parameters in the robot coordinate system. Use the PyRep library to connect the robotic arm, and based on the robotic arm kinematic model, convert the six-degree-of-freedom pose parameters of the target object into robotic arm joint angles. Adopt the RRT (Rapidly-exploring Random Tree) algorithm, combined with the obstacle information preset in the workspace, to plan a collision-free path for the robotic arm from the current position to the target grasping position.

[0078] Control the robotic arm to move along the planned path so that the end effector accurately reaches the grasping position of the target object. First, open the end effector to wrap the target object, and then close the gripper to complete the grasping action. Finally, control the robotic arm to carry the grasped object to the specified position, release the gripper to put down the object, and successfully complete the entire robotic arm grasping operation process.

[0079] In summary, the present invention uses a collaborative software and hardware architecture and phased processing to automate the entire process from image acquisition to precise gripping by the robotic arm: at the hardware level, a perception-execution system for the industrial camera and the UR5e robotic arm is constructed, and at the software level, template matching, pose solution, and path planning are performed based on OpenCV and PyRep. Through multi-angle image acquisition and template creation, a highly robust feature benchmark is provided for target recognition; template matching and threshold screening are used to quickly locate the coordinates of the object in the image, solving the problem of target detection under complex lighting conditions; combining camera calibration parameters with the principle of triangulation, the two-dimensional image coordinates are converted into three-dimensional pose parameters to ensure the robotic arm's accurate perception of the spatial position of the object; the RRT algorithm is used to plan a collision-free path, avoid obstacles in the workspace, and improve movement safety; the precise positioning and adaptive gripping control of the end effector ensure the stable gripping of objects of different regular shapes.

[0080] Next, an image processing device proposed according to an embodiment of the present application is described with reference to the accompanying drawings.

[0081] Figure 7 It is a block diagram of an image processing device according to an embodiment of the present application.

[0082] like Figure 7 As shown, the image processing device 10 includes: an acquisition module 100 , a construction module 200 , a calibration module 300 , a calculation module 400 and a generation module 500 .

[0083] Among them, the acquisition module 100 is used to obtain structured light coded images, polarization imaging images and binocular depth images; the construction module 200 is used to parse the structured light coded images through the structured light decoding algorithm to extract the edge contour and phase information of the target object; the polarization imaging image is analyzed using the polarization calculation model to obtain the surface material and reflective properties of the object; according to the edge contour, phase information and surface material and reflective properties, the three-dimensional coordinate data of the binocular depth image is integrated to construct a multi-dimensional feature vector; the calibration module 300 is used to dynamically calibrate the multi-dimensional feature vector according to the ambient lighting parameters and the robot posture data, and correct the lighting changes through the adaptive white balance algorithm The color deviation caused by the calibration parameters of the robot end effector is used to compensate for the camera installation error, and a calibrated target feature descriptor is generated; the calculation module 400 is used to detect the geometric primitives of the target object based on the calibrated target feature descriptor, use the improved Hough transform to identify the object category and posture in combination with the template matching algorithm, and calculate the six-degree-of-freedom pose parameters of the target object in the robot coordinate system through triangulation; the generation module 500 is used to generate a collision-free grasping path based on the six-degree-of-freedom pose parameters, combined with the robot kinematic model and workspace obstacle information, and control the robot end effector to complete the grasping task of the target object.

[0084] It should be noted that the foregoing explanation of the embodiments of an image processing method also applies to an image processing apparatus in this embodiment, and will not be elaborated here.

[0085] An image processing apparatus according to an embodiment of the present application accurately extracts the edge contour and phase information of a target object through a structured light decoding algorithm, analyzes the surface material and reflection characteristics using a polarization degree calculation model, and constructs a multi-dimensional feature vector including geometric shape, material attributes, and spatial coordinates by combining binocular depth data to describe the multi-dimensional features of the target. For environmental light changes and camera installation errors, the color deviation is dynamically corrected through an adaptive white balance algorithm, and the hardware error is compensated using the calibration parameters of the end effector of the robot to improve the environmental robustness of the feature descriptor. Based on an improved Hough transform and template matching algorithm, combined with triangulation technology, geometric primitives of the target are detected, the object category and pose are recognized, and six-degree-of-freedom pose parameters are calculated, reducing the misjudgment rate of target detection in complex backgrounds by more than 60% and improving the pose calculation accuracy to the millimeter and angle levels. In the path planning stage, based on a high-precision kinematic model, an improved weighted RRTstar algorithm and a dynamic obstacle detection mechanism are integrated to generate a collision-free path that meets kinematic constraints, obstacle avoidance requirements, and grasping stability, shortening the path planning time of the robot in a dense obstacle scenario by 40% and increasing the grasping success rate of the end effector to 96%, enhancing the target recognition accuracy, pose calculation accuracy, and path planning safety of the robot in complex environments. Thus, the problems of insufficient image parsing ability, single compensation mechanism, and lack of path optimization ability in the prior art are solved.

[0086] Figure 8 FIG. 20 is a schematic structural diagram of a robot 20 provided by an embodiment of the present application. The robot 20 may include:

[0087] A robot main body A1 and a moving mechanism A2; among them, the robot main body A1 is a frame structure, and core components such as a power supply module and a data processing module are arranged inside. The power supply module and the data processing module are used to provide electrical energy and core function supports such as data operation and control for the robot, ensuring the stable operation of the overall robot system; the moving mechanism A2 includes a driving motor, a transmission component, and a walking device. The driving motor is used to convert electrical energy into mechanical energy to drive mechanical components, and the transmission component is used to transmit power and motion in the mechanical system, perform energy transfer between the power source and the actuator, and can change the speed, torque, direction, or form of the motion. The walking device is used to select a wheeled, tracked, or legged structure according to the application scenario for movement.

[0088] In an embodiment of the present application, the robot further includes: an image processing system A3 and a robotic arm A4.

[0089] Among them, the image processing system A3 includes a camera, an image acquisition card, and an image processor. The camera is used to collect image information in the working environment. The image acquisition card is used to convert the analog image signal collected by the camera into a digital signal and transmit it to the image processor. The image processor is used to perform real-time processing on the digital image, including operations such as image preprocessing, target detection, feature extraction, and positioning calculation. The robotic arm A4 is a multi-degree-of-freedom mechanical structure. Servo motors for driving joint rotation and angle sensors are provided at each joint of the robotic arm. The servo motors for driving joint rotation are used to precisely control the rotation angle and speed of each joint of the robotic arm, and the angle sensors are used to real-time feedback the position information of the joints to perform precise positioning and motion control of the robotic arm.

[0090] In the embodiment of the present application, the robot further includes: an end effector A5 and a control system A6.

[0091] Among them, the end effector A5 includes a pressure sensor and a tactile sensor. The pressure sensor and the tactile sensor are used to sense the pressure and contact situation when grasping an object, to avoid damaging the object due to excessive grasping force or dropping the object due to insufficient grasping force. The control system A6 includes a hardware controller and control software. The hardware controller is used to receive the target object information transmitted by the image processing system, combine the current position and attitude information of the robotic arm, generate control instructions for each joint of the robotic arm through a motion planning algorithm, and send them to the servo motor to control the motion of the robotic arm, so that the end effector accurately reaches the grasping position of the target object and performs the grasping action. The tactile sensor is used to perform coordinated control on the robot system, communicate with external devices, and send the working status information of the robot.

[0092] It can be understood that the embodiment of the present application performs efficient operations through multi-module collaborative design: the frame-type main body integrates core components to ensure system stability, and the diversified mobile mechanisms (wheel type / crawler type / legged type) adapt to complex scenarios to improve the environmental adaptability; the image processing system combines a camera and an image processor to achieve real-time environment perception and target positioning, providing data support for precise operations; the multi-degree-of-freedom robotic arm is equipped with servo motors and angle sensors, and cooperates with the pressure / tactile sensors of the end effector to precisely control the grasping force and sense the contact state, avoiding object damage or dropping, and ensuring the safety and reliability of operations; the control system integrates target information and the state of the robotic arm through a motion planning algorithm to generate precise control instructions, performing closed-loop intelligent control from environment perception to action execution, supporting external communication and status feedback, and improving the task execution efficiency and intelligent level of the robot in industrial, service and other fields.

[0093] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic expressions of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0094] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0095] Any process or method description shown in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or more executable instructions for implementing a customized logic function or process, and the scope of the preferred embodiments of this application includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in the reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the technical field to which the embodiments of this application belong.

[0096] It should be understood that each part of this application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following well-known technologies in the art or a combination thereof can be used: discrete logic circuits with logic gate circuits for implementing logic functions on data signals, application-specific integrated circuits with suitable combinational logic gate circuits, programmable gate arrays (PGA), field-programmable gate arrays (FPGA), etc.

[0097] Although the embodiments of this application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting this application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. An image processing method, characterized in that, Including: Obtaining a structured light encoded image, a polarization imaging image, and a binocular depth image; Analyzing the structured light encoded image through a structured light decoding algorithm to extract the edge contour and phase information of the target object; Analyzing the polarization imaging image using a polarization degree calculation model to obtain the surface material and reflection characteristics of the object; Based on the edge contour, phase information, surface material, and reflection characteristics, fusing the three-dimensional coordinate data of the binocular depth image to construct a multi-dimensional feature vector; Dynamically calibrating the multi-dimensional feature vector according to the environmental light parameters and the robot pose data, correcting the color deviation caused by light changes through an adaptive white balance algorithm, and compensating for the camera installation error using the calibration parameters of the robot end effector to generate a calibrated target feature descriptor; Based on the calibrated target feature descriptor, using an improved Hough transform to detect the geometric primitives of the target object, combining a template matching algorithm to identify the object category and pose, and calculating the six-degree-of-freedom pose parameters of the target object in the robot coordinate system through triangulation. Among them, the formula for the improved Hough transform is: Among them, is the cumulative function value; are the abscissa and ordinate of the center of the candidate circle; is the radius of the candidate circle; i is the summation index; is the total number of data points; are the abscissa and ordinate of the i-th data point; is the standard deviation; is the natural exponential function; According to the six-degree-of-freedom pose parameters, combining the robot kinematic model and the workspace obstacle information, generating a collision-free grasping path, and controlling the robot end effector to complete the grasping task of the target object.

2. The image processing method according to claim 1, wherein According to the six-degree-of-freedom pose parameters, combining the robot kinematic model and the workspace obstacle information, generating a collision-free grasping path, including: Constructing a high-precision forward and inverse kinematic solution model; Based on the high-precision forward and inverse kinematic solution model, mapping the six-dimensional pose parameters of the target object to the robotic arm joint space, and planning a path in combination with an improved weighted RRTstar algorithm. Among them, the improved weighted RRTstar algorithm in the node expansion process includes: embedding real-time collision detection, quickly detecting collisions with workspace obstacles through a bounding box hierarchy tree; adopting an adaptive step size strategy, dynamically adjusting the expansion step size according to the obstacle density to balance the path length and joint motion smoothness; Based on the planned path, combining the dynamic obstacle map update mechanism of point cloud semantic segmentation and the robotic arm dynamics constraints, constructing a path evaluation function, online optimizing the path, and generating a composite path that meets the kinematic constraints, obstacle avoidance requirements, and grasping pose stability.

3. The image processing method according to claim 1, wherein Analyzing the structured light encoded image through a structured light decoding algorithm to extract the edge contour and phase information of the target object, including: Obtaining a structured light encoded image; Decoding the Gray code of the structured light encoded image, determining the fringe order of each pixel point through row-by-row scanning, and calculating the wrapped phase using the four-step phase-shifting method; Based on the fringe order, performing phase unwrapping, using a quality-guided phase unwrapping algorithm to generate a continuous phase map, and combining the camera-projector calibration parameters to convert the phase value into a three-dimensional coordinate point cloud through a triangulation model; Performing edge extraction on the three-dimensional coordinate point cloud, using the normal change rate detection method, performing non-maximum suppression and double-threshold connection on the edge points to generate a continuous three-dimensional edge contour.

4. The image processing method according to claim 3, wherein The formula for the four-step phase-shifting method is: wherein, is the phase value at a certain point (x, y) on the image or plane; is the arctangent function; are the light intensity or image intensity measurement values under four different phase offsets; is the phase offset 0 is the phase offset 90; is the phase offset 180; is the phase offset 270.

5. The image processing method according to claim 1, wherein Analyzing the polarization imaging image using a polarization degree calculation model, including: Construct a polarization state analysis network; Based on the polarization state analysis network, the polarization degree and polarization angle are calculated by using Stokes parameters; Dynamically adjusting a polarization filter threshold according to the reflective characteristics of the polarization degree and the polarization angle of the target surface; The multi-angle polarization imaging data is fused with the polarization filter threshold to generate a material classification feature map.

6. An image processing apparatus for performing the image processing method according to any one of claims 1 to 5, characterized in that, include: An acquisition module is used to acquire structured light coded images, polarization imaging images, and binocular depth images; A construction module is configured to parse the structured light coded image using a structured light decoding algorithm to extract the edge contour and phase information of the target object; analyze the polarization imaging image using a polarization degree calculation model to obtain the surface material and reflective properties of the object; and fuse the three-dimensional coordinate data of the binocular depth image based on the edge contour, phase information, surface material, and reflective properties to construct a multidimensional feature vector; a calibration module for dynamically calibrating the multidimensional feature vector according to ambient lighting parameters and robot posture data, correcting color deviation caused by lighting changes using an adaptive white balance algorithm, compensating for camera installation errors using calibration parameters of the robot end effector, and generating a calibrated target feature descriptor; The calculation module is used to detect the geometric primitives of the target object using the improved Hough transform based on the calibrated target feature descriptor, identify the object category and posture in combination with the template matching algorithm, and calculate the six-degree-of-freedom pose parameters of the target object in the robot coordinate system by triangulation. The formula of the improved Hough transform is: Among them, is the cumulative function value; are the abscissa and ordinate of the center of the candidate circle; is the radius of the candidate circle; i is the summation index; is the total number of data points; are the abscissa and ordinate of the i-th data point; is the standard deviation; is the natural exponential function; The generation module is used to generate a collision-free grasping path based on the six-degree-of-freedom posture parameters, combined with the robot kinematic model and workspace obstacle information, and control the robot end effector to complete the grasping task of the target object.

Images