Path planning method and system for inspection robot

Abstract

The invention belongs to the technical field of inspection robot systems, and particularly relates to an inspection robot path planning method and system.The visual semantic perception module collects an industrial environment image through a top industrial camera, after graying and Gaussian filtering preprocessing, feature points are detected and matched through an ORB algorithm, and a path planning result is obtained; in combination with an illumination self-adaptive threshold screening mechanism, mismatching points are eliminated, semantics are marked, and a semantic feature map is constructed; the initial path planning module generates an initial path through a semantic cost-containing A * algorithm based on the map; the dynamic obstacle avoidance module captures a moving obstacle by using a visual sensor, and predicts a trajectory through Kalman filtering; the path optimization module combines an initial path and an obstacle track, optimizes the path by using quadratic programming of a fusion curvature constraint and an energy consumption model, and corrects positioning by fusing vision and IMU data through a dynamic weight fusion algorithm; and the execution feedback module generates an instruction according to the optimized path, re-triggers path optimization, forms a closed loop, and ensures the inspection stability.

Description

Technical Field

[0001] This invention belongs to the technical field of inspection robot systems, specifically a path planning method and system for inspection robots. Background Technology

[0002] In the field of industrial automation, inspection robots are core equipment for ensuring stable equipment operation and reducing the risks of manual inspection. The accuracy, safety, and adaptability of their path planning directly determine inspection efficiency and work quality. Current mainstream inspection robot path planning technologies face several limitations in complex industrial environments: Firstly, industrial scenes suffer from large fluctuations in light intensity. Existing visual feature extraction and matching technologies often use fixed thresholds to select feature points, which are easily affected by lighting interference, resulting in numerous mismatched feature points. This leads to insufficient accuracy in the constructed environmental map, creating potential positioning errors for subsequent path planning. Secondly, traditional path planning algorithms (such as the conventional A* algorithm) only use path length as the core optimization objective, failing to fully consider the differences in equipment hazard levels in industrial scenarios. The planned paths may approach high-risk areas, posing safety hazards and failing to meet the core requirement of "safety first" in industrial inspection. Meanwhile, the movement patterns of moving obstacles in industrial scenarios are complex. Existing dynamic obstacle avoidance technologies mostly use general motion models to predict trajectories, without optimizing the models for the characteristics of "intermittent movement and frequent turning" of obstacles in industrial scenarios. This results in low trajectory prediction accuracy and delayed obstacle avoidance response. In addition, the path smoothing optimization stage often ignores the relationship between robot motion energy consumption and mechanical wear. The planned path is prone to problems such as too sharp turns, which not only increases robot energy consumption and shortens endurance, but may also affect driving stability due to frequent turns. The above problems together restrict the reliable application of inspection robots in industrial scenarios. Summary of the Invention

[0003] To address the problems mentioned in the background art, the present invention provides a path planning method and system for inspection robots, thereby solving the problems of low trajectory prediction accuracy and delayed obstacle avoidance response of inspection robots.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a path planning method for an inspection robot, characterized by comprising the following steps: Step S1: Acquire industrial environment images using an industrial camera mounted on top of the inspection robot to ensure complete capture of equipment layout, channel distribution, and potential obstacle information within the industrial environment. For each frame of the acquired industrial environment image, grayscale conversion and noise reduction preprocessing are performed. Gaussian filtering is used to remove salt-and-pepper noise and Gaussian noise from the image. Then, the ORB feature extraction algorithm is applied to detect and match feature points in the preprocessed image. The ORB feature extraction algorithm quickly locates feature points in the image using FAST corner detection and combines BRIEF descriptors to describe the feature points for matching. Based on this, according to the real-time light intensity in the industrial environment, the matched feature points are filtered using the following adaptive threshold filtering mechanism to eliminate mismatched feature points caused by light interference: ; in, Feature point selection and scoring, The gradient value at the feature point is the image gradient value. The larger the gradient value, the higher the contrast of the feature point in the image, and the more accurately it reflects the environmental features. This represents the number of matching points for the feature point in 5 adjacent frames. The more matching points there are, the stronger the stability of the feature point. Step S2: Based on the semantic feature map constructed in Step S1, the A* algorithm is used for initial path planning. The optimal path is found by comprehensively considering the actual cost from the starting point to the current node and the estimated cost from the current node to the target node. The actual cost from the starting point to the current node reflects the length of the traversed path, while the estimated cost from the current node to the target node is calculated based on Euclidean distance or Manhattan distance. This method, based on the A* algorithm, introduces an evaluation function that integrates semantic costs to achieve a balance between path length and industrial safety. The specific evaluation function is as follows: ; in, The actual cost from the starting point to the current node n is obtained by calculating the Euclidean distance of the path between the starting point and the current node. The heuristic function for the current node n to the target node is calculated using Euclidean distance; Assign a semantic cost to the current node n based on the device hazard level corresponding to the location of the node; =0.8 is the semantic weight coefficient, which is used to adjust the degree of influence of semantic cost in the evaluation function; Step S3: For common moving obstacles in industrial scenarios, firstly, the visual sensor installed at the front end of the inspection robot captures real-time image information of the moving obstacles, and combines image recognition technology to determine the type, size, and real-time position of the moving obstacles; then, the Kalman filter algorithm is used to predict the motion trajectory of the moving obstacles in advance to plan obstacle avoidance strategies; a linear piecewise motion model is constructed to describe the motion state of the moving obstacles. This model considers the motion modes that obstacles may exhibit in industrial scenarios, such as uniform speed driving, deceleration and stopping, and turning. ; Wherein, the state vector It includes the x-axis coordinates, y-axis coordinates, moving speed, and heading angle of the moving obstacle at time k; This is the state transition matrix, dynamically adjusted based on the obstacle's motion state at the previous moment. If the obstacle was moving at a constant velocity in a straight line at the previous moment... This is the control matrix, used to introduce control inputs for obstacles; The process noise follows a Gaussian distribution with a mean of 0 and a preset variance, used to describe random disturbances during obstacle movement. The observation equation uses visual flow detection data to obtain the observed position of the obstacle by analyzing the pixel displacement of the obstacle in adjacent frames. ; in, The observation vector at time k contains the observed x-axis and y-axis coordinates of the obstacle; This is the observation matrix, used to map the state vector to the observation space; To observe the noise, it also follows a Gaussian distribution; Step S4: Based on the initial path generated in Step S2 and the obstacle dynamic trajectory predicted in Step S3, perform energy consumption-aware smoothing optimization on the initial path to address potential issues such as sharp turns and high energy consumption, while avoiding dynamic obstacles. First, analyze the motion characteristics of the inspection robot. During movement, changes in linear and angular velocity directly affect energy consumption, and excessive path curvature leads to frequent robot turns, increasing mechanical wear and energy consumption. Therefore, construct a quadratic programming objective function that integrates curvature constraints and an energy consumption model, with the optimization objective being to minimize path curvature and robot energy consumption. ; in, Let be the curvature at the i-th node of the path. The smaller the curvature, the smoother the path. This is the energy consumption model for the robot at the i-th node of the path. Let be the linear velocity of the robot at that node. Angular velocity, and the sum of the squares of linear velocity and angular velocity can effectively reflect the robot's energy consumption level; This is the energy consumption weighting coefficient, used to balance the priority of path smoothness and energy consumption optimization; Step S5: During the inspection robot's journey along the optimized path, a multi-sensor fusion positioning correction scheme is adopted to avoid positioning deviations caused by errors from a single sensor; the visual feature matching confidence level is obtained through the industrial camera from step S1. The confidence level is calculated based on the number of feature matches between the current frame image and the semantic feature map. The more matches, the higher the confidence level, with a value ranging from 0 to 1. Simultaneously, the IMU (Inertial Measurement Unit) confidence level is obtained from the IMU carried by the inspection robot. The IMU confidence score is calculated based on the IMU's drift rate; the lower the drift rate, the higher the confidence score, with a value range of 0-1. Based on this, the following dynamic weight fusion algorithm is used to determine the weight of the visual sensor in the fusion localization: ; Specifically, when the positioning deviation > 0.3m, it indicates that the current positioning accuracy is low, and the weight of the visual sensor needs to be increased to utilize its high-precision feature matching advantage; in this case, α = 0.8 is set. When the positioning deviation ≤ 0.3m, the positioning accuracy meets the requirements, and the weight of the visual sensor is appropriately reduced to decrease the computational load; in this case, α = 0.5 is set. Based on the calculated visual sensor weights... With IMU weights (1- The system fuses visual positioning data with IMU positioning data, and then uses extended Kalman filtering to correct errors in the fused positioning data. This process corrects the robot's position and attitude information in real time, ensuring that the robot's positioning accuracy in the industrial environment is ≤0.1m and avoiding path deviation due to positioning errors. Step S6: Based on the optimal path optimized in Step S4 and combined with the motion control parameters of the inspection robot, generate corresponding speed control commands, including linear velocity and angular velocity commands within each control cycle; execute the speed control commands through a PID controller, which adjusts the control output in real time by comparing the deviation between the robot's actual movement speed and the commanded speed, enabling the robot to accurately track the planned path; during the robot's movement, collect visual loop closure detection data in real time, and determine whether the robot has path deviation by performing feature matching between the current frame image and historical key frame images; when a path deviation is detected, it indicates that the current path can no longer meet the inspection requirements, and immediately re-trigger the path optimization process of Step S4, combining the latest semantic information of the current environment and obstacle dynamic information to regenerate the optimal path, forming a closed-loop control from path planning to optimization, ensuring that the inspection robot continuously and stably completes the industrial inspection task.

[0005] A path planning system for an inspection robot according to claim 1, characterized in that it comprises: The visual semantic perception module consists of an industrial camera and an image processor. It is used to perform the function of step S1 in claim 1. The industrial camera is responsible for acquiring images of the industrial environment, and the image processor is responsible for preprocessing, feature extraction, matching and filtering, and semantic annotation of feature points, and finally outputs a semantic feature map. Initial path planning module: Based on the semantic feature map output by the visual semantic perception module, it executes the function of step S2 in claim 1, runs the A* algorithm and combines it with the evaluation function of fused semantic cost to generate an initial path. Dynamic obstacle avoidance module: integrates a visual sensor, performs the function of step S3 in claim 1, captures information of moving obstacles and uses the Kalman filter algorithm to predict their motion trajectory, and outputs obstacle avoidance decision instructions.

[0006] Path optimization module: Based on the initial path output by the initial path planning module and the obstacle avoidance command output by the dynamic obstacle avoidance module, it executes the functions of steps S4 and S5 in claim 1, optimizes the path smoothness through a quadratic planning algorithm, corrects the positioning error by combining a multi-sensor fusion positioning algorithm, and outputs the optimal control parameters. The execution feedback module includes a motion controller and a feedback sensor. It performs the function of step S6 in claim 1. The motion controller executes speed control commands, and the feedback sensor collects pose data and detects path deviation. When the deviation exceeds the limit, the path is re-optimized.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: In this invention, step S1 employs an adaptive threshold filtering mechanism that adjusts the coefficients α and β based on the real-time light intensity of the industrial environment. This effectively solves the problem of fixed threshold filtering of feature points being easily affected by light interference in existing technologies. When the light is sufficient, the focus is on filtering stable matching feature points; when the light is dim, the focus is on filtering high-contrast feature points. This significantly reduces the false matching rate, improves the accuracy and robustness of feature point filtering, provides reliable data support for the subsequent construction of a high-precision semantic feature map, and ensures the authenticity and completeness of environmental information collection. In this invention, step S2 introduces an evaluation function that incorporates semantic cost into the A* algorithm. This function quantifies the security risks of different device regions through semantic cost s(n), and combines this with semantic weight coefficients. Balancing path length and safety, it avoids the limitation of the A* algorithm, which only pursues the shortest path. The planned initial path can proactively avoid high-risk equipment areas, significantly improving operational safety while ensuring inspection efficiency, directly meeting the core requirement of prioritizing safety in industrial inspections; The linear segmented motion model constructed in step S3 of this invention fully considers the characteristics of "intermittent movement and frequent turning" of moving obstacles in industrial scenarios. By dynamically adjusting the state transition matrix and optimizing the observation equation by combining visual flow detection data, the trajectory prediction accuracy is significantly improved compared with the general motion model. It can accurately predict the movement trajectory of obstacles in advance, leaving sufficient time for the formulation of obstacle avoidance strategies, and solving the problems of poor adaptability and slow response of existing dynamic obstacle avoidance technology in industrial scenarios. In this invention, step S4 is based on a dynamic weight fusion algorithm to adjust the positioning deviation, achieving adaptive fusion of the visual sensor and the IMU, thus overcoming the limitations of existing fixed-weight fusion. When the positioning deviation is large, the visual weight is increased to utilize its high-precision advantage; when the positioning accuracy meets the standard, the visual weight is reduced to reduce the computational load. Combined with extended Kalman filtering to correct the positioning error, the positioning accuracy is effectively reduced, effectively solving the contradiction between positioning accuracy and computational efficiency in complex industrial environments. Attached Figure Description

[0008] Figure 1 This is a flowchart of the overall system in this invention; Figure 2 This is a flowchart of the overall method steps in this invention; Detailed Implementation The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] like Figures 1 to 2 As shown, the present invention provides a path planning method for an inspection robot, characterized by comprising the following steps: Step S1: Acquire industrial environment images using an industrial camera mounted on top of the inspection robot to ensure complete capture of equipment layout, channel distribution, and potential obstacle information within the industrial environment. For each frame of the acquired industrial environment image, grayscale conversion and noise reduction preprocessing are performed. Gaussian filtering is used to remove salt-and-pepper noise and Gaussian noise from the image. Then, the ORB feature extraction algorithm is applied to detect and match feature points in the preprocessed image. The ORB feature extraction algorithm quickly locates feature points in the image using FAST corner detection and combines BRIEF descriptors to describe the feature points for matching. Based on this, according to the real-time light intensity in the industrial environment, the matched feature points are filtered using the following adaptive threshold filtering mechanism to eliminate mismatched feature points caused by light interference: ; in, Feature point selection and scoring, The gradient value at the feature point is the image gradient value. The larger the gradient value, the higher the contrast of the feature point in the image, and the more accurately it reflects the environmental features. This represents the number of matching points for the feature point in 5 adjacent frames. The more matching points there are, the stronger the stability of the feature point. Specifically, step S1 in this invention effectively solves the problem of light interference in the fixed threshold selection of feature points in the prior art by adjusting the coefficients α and β according to the real-time light intensity of the industrial environment through an adaptive threshold selection mechanism. When the light is sufficient, the focus is on selecting stable matching feature points, and when the light is dim, the focus is on selecting high-contrast feature points, which significantly reduces the false matching rate, improves the accuracy and robustness of feature point selection, provides reliable data support for the subsequent construction of a high-precision semantic feature map, and ensures the authenticity and completeness of environmental information collection.

[0010] Specifically, the data acquisition range covers an industrial scene area with a 120°-180° field of view in front of the robot and a distance of 0-10m. The image resolution is set to 1920×1080 pixels, and the acquisition frequency is 15-20 frames / second. When the light intensity is >500 lux, the ambient light is sufficient and the image contrast is high. In this case, α=0.3 and β=0.7 are set, focusing on selecting stable feature points by matching the number of points. When the light intensity is ≤500 lux, the ambient light is dark and the image contrast is reduced. In this case, α=0.6 and β=0.4 are set, focusing on selecting high-contrast feature points by gradient values. For the selected feature points, semantic annotation is performed in conjunction with the pre-set equipment semantic database of the industrial scene. The annotation content includes the equipment type and equipment hazard level corresponding to the feature point. Finally, a semantic feature map containing equipment type, hazard level and spatial coordinate information is constructed to provide accurate environmental semantic basis for subsequent path planning.

[0011] Step S2: Based on the semantic feature map constructed in Step S1, the A* algorithm is used for initial path planning. The optimal path is found by comprehensively considering the actual cost from the starting point to the current node and the estimated cost from the current node to the target node. The actual cost from the starting point to the current node reflects the length of the traversed path, while the estimated cost from the current node to the target node is calculated based on Euclidean distance or Manhattan distance. This method, based on the A* algorithm, introduces an evaluation function that integrates semantic costs to achieve a balance between path length and industrial safety. The specific evaluation function is as follows: ; in, The actual cost from the starting point to the current node n is obtained by calculating the Euclidean distance of the path between the starting point and the current node. The heuristic function for the current node n to the target node is calculated using Euclidean distance; Assign a semantic cost to the current node n based on the device hazard level corresponding to the location of the node; =0.8 is the semantic weight coefficient, which is used to adjust the degree of influence of semantic cost in the evaluation function; Specifically, in step S2 of this invention, an evaluation function incorporating semantic cost is introduced into the A* algorithm. The semantic cost s(n) quantifies the safety risks of different equipment areas, and the semantic weight coefficient balances path length and safety, avoiding the limitation of the A* algorithm which only pursues the shortest path. The planned initial path can proactively avoid high-risk equipment areas, significantly improving operational safety while ensuring inspection efficiency, directly meeting the core requirement of prioritizing safety in industrial inspections.

[0012] Specifically, s(n) represents the semantic cost corresponding to the current node n, which is assigned a value based on the hazard level of the equipment at the location of the node. Areas with high-risk equipment (such as high-voltage distribution cabinets and high-temperature reactors) are assigned a value of 5, areas with general equipment (such as ordinary motors and conveying pipelines) are assigned a value of 2, and open passage areas are assigned a value of 1. The semantic cost reflects the safety risks of different areas. λ=0.8 is the semantic weight coefficient, used to adjust the influence of the semantic cost in the evaluation function. During the path search process, the A* algorithm calculates the score of each candidate node based on the above evaluation function, and prioritizes the node with the lowest score as the next path node. Finally, an initial path is generated from the starting point of the inspection robot to the inspection target point. This initial path can ensure a shorter path length and prioritize avoiding high-risk equipment areas, thereby improving inspection safety.

[0013] Step S3: For common moving obstacles in industrial scenarios, firstly, the visual sensor installed at the front end of the inspection robot captures real-time image information of the moving obstacles, and combines image recognition technology to determine the type, size, and real-time position of the moving obstacles; then, the Kalman filter algorithm is used to predict the motion trajectory of the moving obstacles in advance to plan obstacle avoidance strategies; a linear piecewise motion model is constructed to describe the motion state of the moving obstacles. This model considers the motion modes that obstacles may exhibit in industrial scenarios, such as uniform speed driving, deceleration and stopping, and turning. ; Wherein, the state vector It includes the x-axis coordinates, y-axis coordinates, moving speed, and heading angle of the moving obstacle at time k; This is the state transition matrix, dynamically adjusted based on the obstacle's motion state at the previous moment. If the obstacle was moving at a constant velocity in a straight line at the previous moment... This is the control matrix, used to introduce control inputs for obstacles; The process noise follows a Gaussian distribution with a mean of 0 and a preset variance, used to describe random disturbances during obstacle movement. The observation equation uses visual flow detection data to obtain the observed position of the obstacle by analyzing the pixel displacement of the obstacle in adjacent frames. ; in, The observation vector at time k contains the observed x-axis and y-axis coordinates of the obstacle; This is the observation matrix, used to map the state vector to the observation space; To observe the noise, it also follows a Gaussian distribution; Specifically, the linear segmented motion model constructed in step S3 of this invention fully considers the characteristics of "intermittent movement and frequent turning" of moving obstacles in industrial scenarios. By dynamically adjusting the state transition matrix and optimizing the observation equation by combining visual flow detection data, the trajectory prediction accuracy is significantly improved compared with the general motion model. It can accurately predict the movement trajectory of obstacles in advance, leaving sufficient time for the formulation of obstacle avoidance strategies, and solving the problems of poor adaptability and slow response of existing dynamic obstacle avoidance technologies in industrial scenarios.

[0014] Specifically, the state of the Kalman filter algorithm is updated with a period of 50ms. The motion state estimation of the obstacle is continuously corrected through the prediction-update loop, so as to achieve accurate prediction of the obstacle's trajectory in the next 2 seconds and provide dynamic information of the obstacle for subsequent path optimization.

[0015] Step S4: Based on the initial path generated in Step S2 and the obstacle dynamic trajectory predicted in Step S3, perform energy consumption-aware smoothing optimization on the initial path to address potential issues such as sharp turns and high energy consumption, while avoiding dynamic obstacles. First, analyze the motion characteristics of the inspection robot. During movement, changes in linear and angular velocity directly affect energy consumption, and excessive path curvature leads to frequent robot turns, increasing mechanical wear and energy consumption. Therefore, construct a quadratic programming objective function that integrates curvature constraints and an energy consumption model, with the optimization objective being to minimize path curvature and robot energy consumption. ; in, Let be the curvature at the i-th node of the path. The smaller the curvature, the smoother the path. This is the energy consumption model for the robot at the i-th node of the path. Let be the linear velocity of the robot at that node. Angular velocity, and the sum of the squares of linear velocity and angular velocity can effectively reflect the robot's energy consumption level; This is the energy consumption weighting coefficient, used to balance the priority of path smoothness and energy consumption optimization; Specifically, step S4 in this invention uses a dynamic weight adjustment fusion algorithm based on positioning deviation to achieve adaptive fusion of the visual sensor and the IMU, overcoming the limitations of existing fixed-weight fusion. When the positioning deviation is large, the visual weight is increased to utilize its high-precision advantage; when the positioning accuracy meets the standard, the visual weight is reduced to reduce the computational load. Combined with extended Kalman filtering to correct the positioning error, the positioning accuracy is effectively reduced, effectively solving the contradiction between positioning accuracy and computational efficiency in complex industrial environments.

[0016] Specifically, constraints are set, including that the distance between each node on the path and static equipment and dynamic obstacles must be greater than 0.5m to ensure driving safety, and the rate of curvature change between adjacent nodes on the path must be less than 0.2rad / m to avoid instability caused by the robot turning too fast. The constrained optimization problem is transformed into an unconstrained optimization problem using the Lagrange multiplier method, and the optimal smooth path that satisfies the constraints and minimizes the objective function is obtained. This path ensures safety while taking into account the robot's driving stability and low energy consumption.

[0017] Step S5: During the inspection robot's journey along the optimized path, a multi-sensor fusion positioning correction scheme is adopted to avoid positioning deviations caused by errors from a single sensor; the visual feature matching confidence level is obtained through the industrial camera from step S1. The confidence level is calculated based on the number of feature matches between the current frame image and the semantic feature map. The more matches, the higher the confidence level, with a value ranging from 0 to 1. Simultaneously, the IMU (Inertial Measurement Unit) confidence level is obtained from the IMU carried by the inspection robot. The IMU confidence score is calculated based on the IMU's drift rate; the lower the drift rate, the higher the confidence score, with a value range of 0-1. Based on this, the following dynamic weight fusion algorithm is used to determine the weight of the visual sensor in the fusion localization: ; Specifically, when the positioning deviation > 0.3m, it indicates that the current positioning accuracy is low, and the weight of the visual sensor needs to be increased to utilize its high-precision feature matching advantage; in this case, α = 0.8 is set. When the positioning deviation ≤ 0.3m, the positioning accuracy meets the requirements, and the weight of the visual sensor is appropriately reduced to decrease the computational load; in this case, α = 0.5 is set. Based on the calculated visual sensor weights... With IMU weights (1- The system fuses visual positioning data with IMU positioning data, and then uses extended Kalman filtering to correct errors in the fused positioning data. This process corrects the robot's position and attitude information in real time, ensuring that the robot's positioning accuracy in the industrial environment is ≤0.1m and avoiding path deviation due to positioning errors. Step S6: Based on the optimal path optimized in Step S4 and combined with the motion control parameters of the inspection robot, generate corresponding speed control commands, including linear velocity and angular velocity commands within each control cycle; execute the speed control commands through a PID controller, which adjusts the control output in real time by comparing the deviation between the robot's actual movement speed and the commanded speed, enabling the robot to accurately track the planned path; during the robot's movement, collect visual loop closure detection data in real time, and determine whether the robot has path deviation by performing feature matching between the current frame image and historical key frame images; when a path deviation is detected, it indicates that the current path can no longer meet the inspection requirements, and immediately re-trigger the path optimization process of Step S4, combining the latest semantic information of the current environment and obstacle dynamic information to regenerate the optimal path, forming a closed-loop control from path planning to optimization, ensuring that the inspection robot continuously and stably completes the industrial inspection task.

[0018] A path planning system for an inspection robot according to claim 1, characterized in that it comprises: The visual semantic perception module consists of an industrial camera and an image processor. It is used to perform the function of step S1 in claim 1. The industrial camera is responsible for acquiring images of the industrial environment, and the image processor is responsible for preprocessing, feature extraction, matching and filtering, and semantic annotation of feature points, and finally outputs a semantic feature map. Initial path planning module: Based on the semantic feature map output by the visual semantic perception module, it executes the function of step S2 in claim 1, runs the A* algorithm and combines it with the evaluation function of fused semantic cost to generate an initial path. Dynamic obstacle avoidance module: integrates a visual sensor, performs the function of step S3 in claim 1, captures information of moving obstacles and uses the Kalman filter algorithm to predict their motion trajectory, and outputs obstacle avoidance decision instructions.

[0019] Path optimization module: Based on the initial path output by the initial path planning module and the obstacle avoidance command output by the dynamic obstacle avoidance module, it executes the functions of steps S4 and S5 in claim 1, optimizes the path smoothness through a quadratic planning algorithm, corrects the positioning error by combining a multi-sensor fusion positioning algorithm, and outputs the optimal control parameters. The execution feedback module includes a motion controller and a feedback sensor. It performs the function of step S6 in claim 1. The motion controller executes speed control commands, and the feedback sensor collects pose data and detects path deviations. When the deviation exceeds the limit, path re-optimization is triggered. The working principle and usage process of this invention are as follows: The visual semantic perception module acquires industrial environment images through a top industrial camera. After grayscale conversion and Gaussian filtering preprocessing, it uses the ORB algorithm to detect matching feature points. Combined with an illumination adaptive threshold filtering mechanism, it eliminates mismatched points and labels them semantically, constructing a semantic feature map. The initial path planning module generates an initial path based on this map using the A* algorithm with semantic cost. The dynamic obstacle avoidance module uses a visual sensor to capture moving obstacles and predicts their trajectories using Kalman filtering. The path optimization module combines the initial path and obstacle trajectories and optimizes the path using quadratic planning that integrates curvature constraints and energy consumption models. At the same time, it uses a dynamic weight fusion algorithm to fuse visual and IMU data to correct the positioning. The execution feedback module generates instructions according to the optimized path and controls the robot's movement via PID control. When the visual loop detects a deviation > 0.2m, it re-triggers path optimization to form a closed loop, ensuring stable inspection.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0021] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A path planning method for an inspection robot, characterized in that, Includes the following steps: Step S1: Acquire industrial environment images using an industrial camera mounted on top of the inspection robot to ensure complete capture of equipment layout, channel distribution, and potential obstacle information within the industrial environment. For each frame of the acquired industrial environment image, grayscale conversion and noise reduction preprocessing are performed. Gaussian filtering is used to remove salt-and-pepper noise and Gaussian noise from the image. Then, the ORB feature extraction algorithm is applied to detect and match feature points in the preprocessed image. The ORB feature extraction algorithm quickly locates feature points in the image using FAST corner detection and combines BRIEF descriptors to describe the feature points for matching. Based on this, according to the real-time light intensity in the industrial environment, the matched feature points are filtered using the following adaptive threshold filtering mechanism to eliminate mismatched feature points caused by light interference: ; in, Feature point selection and scoring, The gradient value at the feature point is the image gradient value. The larger the gradient value, the higher the contrast of the feature point in the image, and the more accurately it reflects the environmental features. This represents the number of matching points for the feature point in 5 adjacent frames. The more matching points there are, the stronger the stability of the feature point. Step S2: Based on the semantic feature map constructed in Step S1, the A* algorithm is used for initial path planning. The optimal path is found by comprehensively considering the actual cost from the starting point to the current node and the estimated cost from the current node to the target node. The actual cost from the starting point to the current node reflects the length of the traversed path, while the estimated cost from the current node to the target node is calculated based on Euclidean distance or Manhattan distance. This method, based on the A* algorithm, introduces an evaluation function that integrates semantic costs to achieve a balance between path length and industrial safety. The specific evaluation function is as follows: ; in, The actual cost from the starting point to the current node n is obtained by calculating the Euclidean distance of the path between the starting point and the current node. The heuristic function for the current node n to the target node is calculated using Euclidean distance; Assign a semantic cost to the current node n based on the device hazard level corresponding to the location of the node; =0.8 is the semantic weight coefficient, which is used to adjust the degree of influence of semantic cost in the evaluation function; Step S3: For common moving obstacles in industrial scenarios, firstly, the visual sensor installed at the front end of the inspection robot captures real-time image information of the moving obstacles, and combines image recognition technology to determine the type, size, and real-time position of the moving obstacles; then, the Kalman filter algorithm is used to predict the motion trajectory of the moving obstacles in advance to plan obstacle avoidance strategies; a linear piecewise motion model is constructed to describe the motion state of the moving obstacles. This model considers the motion modes that obstacles may exhibit in industrial scenarios, such as uniform speed driving, deceleration and stopping, and turning. ; Wherein, the state vector It includes the x-axis coordinates, y-axis coordinates, moving speed, and heading angle of the moving obstacle at time k; This is the state transition matrix, dynamically adjusted based on the obstacle's motion state at the previous moment. If the obstacle was moving at a constant velocity in a straight line at the previous moment... This is the control matrix, used to introduce control inputs for obstacles; The process noise follows a Gaussian distribution with a mean of 0 and a preset variance, used to describe random disturbances during obstacle movement. The observation equation uses visual flow detection data to obtain the observed position of the obstacle by analyzing the pixel displacement of the obstacle in adjacent frames. ; in, The observation vector at time k contains the observed x-axis and y-axis coordinates of the obstacle; This is the observation matrix, used to map the state vector to the observation space; To observe the noise, it also follows a Gaussian distribution; Step S4: Based on the initial path generated in Step S2 and the obstacle dynamic trajectory predicted in Step S3, perform energy consumption-aware smoothing optimization on the initial path to address potential issues such as sharp turns and high energy consumption, while avoiding dynamic obstacles. First, analyze the motion characteristics of the inspection robot. During movement, changes in linear and angular velocity directly affect energy consumption, and excessive path curvature leads to frequent robot turns, increasing mechanical wear and energy consumption. Therefore, construct a quadratic programming objective function that integrates curvature constraints and an energy consumption model, with the optimization objective being to minimize path curvature and robot energy consumption. ; in, Let be the curvature at the i-th node of the path. The smaller the curvature, the smoother the path. This is the energy consumption model for the robot at the i-th node of the path. Let be the linear velocity of the robot at that node. Angular velocity, the sum of the squares of linear velocity and angular velocity can effectively reflect the robot's energy consumption level; This is the energy consumption weighting coefficient, used to balance the priority of path smoothness and energy consumption optimization; Step S5: During the inspection robot's journey along the optimized path, a multi-sensor fusion positioning correction scheme is adopted to avoid positioning deviations caused by errors from a single sensor; the visual feature matching confidence level is obtained through the industrial camera from step S1. The confidence level is calculated based on the number of feature matches between the current frame image and the semantic feature map. The more matches, the higher the confidence level, with a value ranging from 0 to 1. Simultaneously, the IMU (Inertial Measurement Unit) confidence level is obtained from the IMU carried by the inspection robot. The IMU confidence score is calculated based on the IMU's drift rate; the lower the drift rate, the higher the confidence score, with a value range of 0-1. Based on this, the following dynamic weight fusion algorithm is used to determine the weight of the visual sensor in the fusion localization: ; Specifically, when the positioning deviation > 0.3m, it indicates that the current positioning accuracy is low, and the weight of the visual sensor needs to be increased to utilize its high-precision feature matching advantage; in this case, α = 0.8 is set. When the positioning deviation ≤ 0.3m, the positioning accuracy meets the requirements, and the weight of the visual sensor is appropriately reduced to decrease the computational load; in this case, α = 0.5 is set. Based on the calculated visual sensor weights... With IMU weights (1- The system fuses visual positioning data with IMU positioning data, and then uses extended Kalman filtering to correct errors in the fused positioning data. This process corrects the robot's position and attitude information in real time, ensuring that the robot's positioning accuracy in the industrial environment is ≤0.1m and avoiding path deviation due to positioning errors. Step S6: Based on the optimal path optimized in Step S4 and combined with the motion control parameters of the inspection robot, generate corresponding speed control commands, including linear velocity and angular velocity commands within each control cycle; execute the speed control commands through a PID controller, which adjusts the control output in real time by comparing the deviation between the robot's actual movement speed and the commanded speed, enabling the robot to accurately track the planned path; during the robot's movement, collect visual loop closure detection data in real time, and determine whether the robot has path deviation by performing feature matching between the current frame image and historical key frame images; when a path deviation is detected, it indicates that the current path can no longer meet the inspection requirements, and immediately re-trigger the path optimization process of Step S4, combining the latest semantic information of the current environment and obstacle dynamic information to regenerate the optimal path, forming a closed-loop control from path planning to optimization, ensuring that the inspection robot continuously and stably completes the industrial inspection task.

2. The inspection robot path planning system according to claim 1, characterized in that, include: The visual semantic perception module consists of an industrial camera and an image processor. It is used to perform the function of step S1 in claim 1. The industrial camera is responsible for acquiring images of the industrial environment, and the image processor is responsible for preprocessing, feature extraction, matching and filtering, and semantic annotation of feature points, and finally outputs a semantic feature map. Initial path planning module: Based on the semantic feature map output by the visual semantic perception module, it executes the function of step S2 in claim 1, runs the A* algorithm and combines it with the evaluation function of fused semantic cost to generate an initial path. Dynamic obstacle avoidance module: integrates a visual sensor, performs the function of step S3 in claim 1, captures information of moving obstacles and uses the Kalman filter algorithm to predict their motion trajectory, and outputs obstacle avoidance decision instructions. Path optimization module: Based on the initial path output by the initial path planning module and the obstacle avoidance command output by the dynamic obstacle avoidance module, it executes the functions of steps S4 and S5 in claim 1, optimizes the path smoothness through a quadratic planning algorithm, corrects the positioning error by combining a multi-sensor fusion positioning algorithm, and outputs the optimal control parameters. The execution feedback module includes a motion controller and a feedback sensor. It performs the function of step S6 in claim 1. The motion controller executes speed control commands, and the feedback sensor collects pose data and detects path deviation. When the deviation exceeds the limit, the path is re-optimized.

Images