Multi-camera coordinated car washing robot path planning system

Abstract

This invention relates to the field of computer vision technology, specifically to a multi-camera collaborative car wash robot path planning system. The system includes a vision acquisition module, an image processing module, a contour extraction module, a path planning module, and a dynamic control module. In this invention, multi-angle image acquisition combined with differentiated exposure and pose recording enables complete acquisition of the vehicle contour and unified posture. Pixel-level distortion correction and brightness contrast normalization improve image quality and contour accuracy. Furthermore, edge enhancement and precise sampling of curvature regions enhance the continuity of surface detail recognition. Path point density matching motion resolution avoids the risk of path overlap with the vehicle body. Simultaneously, real-time error recognition and velocity profile interpolation adjustment ensure smooth output of motion commands and closed-loop control of error response during path following, thereby improving the rationality of path generation and control stability under complex contours.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a multi-camera collaborative car wash robot path planning system. Background Technology

[0002] Computer vision technology involves analyzing and understanding images and videos using computers, enabling machines to recognize, track, and judge target objects in visual scenes. This technology encompasses core aspects such as image acquisition, image preprocessing, feature extraction, target detection, target recognition, and tracking. It has achieved intelligent processing capabilities for visual recognition and scene understanding across multiple industries. Based on visual information acquired by cameras or sensors, this field combines algorithms to understand and judge content in static images or dynamic videos, finding wide applications in traffic monitoring, industrial inspection, medical image analysis, and service robots. Among these applications, the traditional car wash robot path planning system refers to a path generation system designed to enable car wash robots to automatically plan cleaning routes according to vehicle contours. The key technical issue addressed is how to construct a bypass path based on the external structural information of the target vehicle and enable the robot to operate along that path. Traditional car wash robot path planning uses visual image acquisition to identify the vehicle's edge contours, then calculates the path point sequence through contour boundary coordinate extraction, and finally generates an overall path map using coordinate sorting and path connection rules, allowing the car wash robot to move along the vehicle boundaries.

[0003] Existing technologies rely solely on a single visual image for vehicle edge contour recognition. However, limitations caused by fixed-angle image acquisition, such as occlusion and insufficient exposure, can lead to incomplete information or boundary misjudgment during contour extraction. Furthermore, the lack of enhanced recognition of boundary curvature or continuity features during contour point extraction results in sparse boundary distribution in high-curvature areas and reduced path fitting accuracy. Ignoring the minimum working distance and dynamic response capability between the robot and the vehicle boundary during path generation can easily cause path overlap or operation deviation. For example, when dealing with complex contour models such as large SUVs, the robot is prone to steering lag or edge-fitting errors, affecting the integrity of the cleaning operation coverage and the stability of path following. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a multi-camera collaborative car wash robot path planning system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a multi-camera collaborative car wash robot path planning system includes: The vision acquisition module uses multiple sets of cameras distributed on the top and sides of the robot to simultaneously acquire multi-angle images of the vehicle, record timestamps and corresponding poses, perform pixel-level distortion correction and viewpoint correction, identify the vehicle's planar contour information, and generate a set of vehicle spatial contours. The image processing module performs pixel-level noise suppression, differential scale detail adjustment, edge enhancement and continuity restoration, and merges contour refinement and connected components based on the vehicle spatial contour set to generate a vehicle contour image set. The contour extraction module calls the vehicle contour image set, detects the outer boundary pixels of the vehicle, performs fine localization of boundary points, samples and sorts boundary points, removes isolated noise points, identifies the curvature and normal of boundary points, segments straight lines and curves, densely samples high curvature areas, and generates a vehicle contour boundary set. Based on the vehicle contour boundary set, the path planning module sets the robot's working distance and lateral margin, identifies the offset curve on the outside of the vehicle body, samples path points along the offset curve by arc length, adjusts the path point spacing to match the robot's motion resolution, and removes points that overlap with the vehicle contour to obtain the vehicle-around path planning table.

[0006] As a further embodiment of the present invention, the vehicle spatial contour set includes vehicle planar contour information, boundary point distribution features, and corresponding distance evaluation values; the vehicle contour image set includes denoised images, edge-enhanced images, and boundary continuity evaluation values; the vehicle contour boundary set includes boundary point sequences, boundary distribution features, and boundary continuity indices; and the vehicle-around path planning table includes path point sequences, path offsets, and path adjustment intervals.

[0007] As a further aspect of the present invention, the visual acquisition module includes: The multi-view acquisition submodule uses multiple cameras on the top and sides of the robot to perform multi-camera synchronous startup and exposure control, acquire multi-frame images of the vehicle from different angles, record the timestamp of each frame and camera pose parameters, perform pixel-level distortion correction and image registration, and use inverse perspective transformation to convert the image into a planar view to generate vehicle planar contour information. The ranging evaluation submodule calls the vehicle's planar contour information, sets multiple reference ranging points on the image width and height, calibrates the pixel coordinates corresponding to each ranging point, extracts pixel grayscale or color features for boundary recognition, calculates the pixel distance from the ranging point to the vehicle boundary, converts it into the spatial distance between the robot and the vehicle, and generates the corresponding distance evaluation value. The spatial information integration submodule calls the corresponding distance evaluation value, combines it with the vehicle planar contour information, performs coordinate mapping on the spatial position of the ranging point, merges adjacent or overlapping ranging results, removes abnormal deviation data, and outputs the vehicle spatial contour set.

[0008] As a further aspect of the present invention, the image processing module includes: The denoising submodule, based on the vehicle spatial contour set, locates high-noise areas in the image, separates the background and target areas, performs pixel-level noise point removal, smooths local color and brightness changes, repairs disturbed contour segments, and generates a denoised image. The edge enhancement submodule calls the denoised image, detects the vehicle boundary region, separates boundary pixels from non-boundary pixels, optimizes blurred edges, connects broken contour lines, balances boundary grayscale distribution, and generates an edge-enhanced image. The boundary continuity assessment submodule calls the edge enhancement image, extracts the complete boundary pixel set, arranges the boundary points in spatial order, calculates the distance difference between adjacent boundary points, marks missing or abrupt boundary segments, calculates the proportion of continuous and discontinuous segments, and generates a vehicle contour image set.

[0009] As a further aspect of the present invention, the contour extraction module includes: The boundary point recognition submodule locates the area where the vehicle's outer boundary is located based on the vehicle contour image set, separates the boundary area from the non-boundary area, scans the pixels of the boundary area point by point, records the pixel coordinates that conform to the outer boundary features, sorts the boundary points in spatial order, and obtains the boundary point distribution features. The boundary distribution analysis submodule calls the boundary point distribution characteristics, calculates the distribution density of boundary points in the horizontal and vertical directions, counts the range of variation in the distance between boundary points, detects the location of dense and sparse areas, analyzes the changes in the direction of boundary points, and obtains boundary distribution characteristic data. The contour integrity assessment submodule calls the boundary distribution feature data, reconstructs the boundary point connection to form a closed contour, detects the length and position of contour gaps, calculates the ratio of continuous segments to broken segments, measures the smoothness of contour lines, marks missing boundary segments, and generates a vehicle contour boundary set.

[0010] As a further aspect of the present invention, the path planning module includes: The path point calculation submodule extracts the spatial coordinates of boundary points based on the vehicle outline boundary set, generates path reference points along the vehicle outline at fixed intervals, identifies the tangential and normal directions of the path reference points, corrects points that do not match the vehicle outline, connects the path reference points to form a continuous path line, and obtains the path point distribution information. The path offset analysis submodule calls the path point distribution information, analyzes the displacement difference between adjacent path points in the horizontal and vertical directions, counts the average offset of continuous point segments, calculates the average offset value of continuous point segments, detects offset abrupt change points and records their positions, analyzes the distance change between the path line and the vehicle outline, and generates the path offset. The path adjustment interval determination submodule calls the path offset, combines the vehicle shape change characteristics, locates the start and end positions of the path segment that needs to be adjusted, sets the correction direction and correction magnitude of the adjusted segment, integrates the adjusted segment and the unadjusted segment, and generates a vehicle-around path planning table.

[0011] As a further aspect of the present invention, the system also includes a dynamic control module: The dynamic control module calls the vehicle path planning table, estimates the path index corresponding to the current pose, identifies lateral and heading errors, combines the velocity profile, calculates the desired speed and heading by interpolation, applies acceleration / deceleration and steering angular velocity limits, outputs servo and drive control commands, monitors execution errors, marks trajectory deviations, and obtains the dynamic control signal set of the car wash robot. The dynamic control signal set of the car wash robot includes speed adjustment commands, steering angle adjustment commands, and execution time synchronization information.

[0012] As a further aspect of the present invention, the dynamic control module includes: Based on the vehicle path planning table, the speed adjustment submodule extracts the robot's current position coordinates, matches the reference points in the path point sequence from the current position, identifies the distance difference between the current position and the next reference point, determines the speed change range by combining the path curvature change, and generates a speed adjustment instruction set. The steering angle adjustment submodule calls the speed adjustment instruction set, combines the path point sequence and path offset, identifies the turning radius from the current path point to the next path point, determines the steering direction and angle range, corrects the steering angle under the influence of the offset, and generates the steering angle adjustment instruction set. The execution time synchronization submodule calls the steering angle adjustment instruction set, extracts the timestamps of the speed adjustment and steering angle adjustment instructions, calculates the time difference between the two, adjusts the execution order of the instructions, synchronizes the corresponding execution start and end times, and combines the synchronized speed and steering instructions to obtain the dynamic control signal set of the car wash robot.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, multi-angle image acquisition combined with differentiated exposure and pose recording methods enables complete acquisition of vehicle contours and unified posture processing. Pixel-level distortion correction and brightness contrast normalization improve image quality and contour accuracy. Edge enhancement and precise sampling of curvature regions further enhance the continuity and completeness of surface detail recognition. Matching path point density to actual motion resolution effectively avoids the risk of path-vehicle overlap. Real-time error recognition and velocity profile interpolation adjustment ensure smooth motion commands and closed-loop error response during path following, thereby improving the rationality of path generation and dynamic control accuracy in complex contour environments, and enhancing path planning and execution stability. Attached Figure Description

[0014] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the visual acquisition module in this invention; Figure 3 This is a flowchart of the image processing module in this invention; Figure 4 This is a flowchart of the contour extraction module in this invention; Figure 5 This is a flowchart of the path planning module in this invention; Figure 6 This is a flowchart of the dynamic control module in this invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0017] Please see Figure 1 The multi-camera collaborative car wash robot path planning system includes: The vision acquisition module uses multiple sets of cameras distributed on the top and sides of the robot to simultaneously acquire multi-angle images of the vehicle, capture frames with different exposure levels, record timestamps and corresponding poses, perform pixel-level distortion correction and viewpoint correction, identify vehicle planar contour information, and generate a vehicle spatial contour set. The image processing module performs pixel-level noise suppression, differential scale detail adjustment, edge enhancement and continuity restoration based on the vehicle spatial contour set, normalizes brightness and contrast, merges contour thinning and connected components, and generates a vehicle contour image set. The contour extraction module calls the vehicle contour image set, detects the pixels of the vehicle's outer boundary, performs fine-grained boundary point localization, samples and sorts the boundary points, removes isolated noise points, identifies the curvature and normal of the boundary points, segments straight lines and curves, densely samples high curvature areas, and generates the vehicle contour boundary set. The path planning module is based on the vehicle contour boundary set, sets the robot operation spacing and lateral margin, identifies the offset curve on the outside of the vehicle body, samples path points along the offset curve according to the arc length, adjusts the path point spacing to match the robot motion resolution, and removes points that overlap with the vehicle contour to obtain the vehicle-around path planning table. The dynamic control module calls the vehicle path planning table, estimates the path index corresponding to the current pose, identifies lateral and heading errors, combines the velocity profile, calculates the desired speed and heading through interpolation, applies acceleration / deceleration and steering angular velocity limits, outputs servo and drive control commands, monitors execution errors, marks trajectory deviations, and obtains the dynamic control signal set of the car wash robot.

[0018] The vehicle spatial contour set includes vehicle planar contour information, boundary point distribution characteristics, and corresponding distance evaluation values. The vehicle contour image set includes denoised images, edge-enhanced images, and boundary continuity evaluation values. The vehicle contour boundary set includes boundary point sequences, boundary distribution characteristics, and boundary continuity indices. The vehicle path planning table includes path point sequences, path offsets, and path adjustment intervals. The car wash robot dynamic control signal set includes speed adjustment commands, steering angle adjustment commands, and execution time synchronization information.

[0019] Please see Figure 2 The visual acquisition module includes: The multi-view acquisition submodule uses multiple cameras on the top and sides of the robot to perform multi-camera synchronous startup and exposure control, acquire multi-frame images of the vehicle from different angles, record the timestamp of each frame and camera pose parameters, perform pixel-level distortion correction and image registration, and use inverse perspective transformation to convert the image into a planar view to generate vehicle planar contour information. Multiple cameras, such as Baumer TXG50 and Baslerac A2040-35gc, are used on the top and sides of the robot. The camera array has a horizontal field of view of 270 degrees. The controller sends a 5V TTL signal to synchronously start the cameras. Based on 500 lux illumination, the exposure time is set to 10 milliseconds. The cameras acquire multiple frames of images of the vehicle at 10 frames per second. For example, one frame is acquired every 30 centimeters the robot moves, for a total of 120 frames. Each frame is recorded with a microsecond-level timestamp. Internal parameters such as focal length and principal point coordinates, as well as camera pose parameters (rotation matrix R1, translation vector T1), are obtained through camera calibration. Parameters such as k1=0.05 and p1=0.005 are used. The distortion coefficient corrects barrel distortion in the image, extracts SIFT feature points, matches at least 200 pairs of feature points, and uses the RANSAC algorithm to solve the homography matrix H, for example, calculating H_top1_side1 between the top camera and the side camera 1 image. The image is registered to the robot base coordinate system, and the ground plane at the bottom of the vehicle is selected as the projection reference. The inverse perspective transformation matrix P is constructed. For example, by combining the camera intrinsic parameters, extrinsic parameters and ground equations, the image points (u, v) are converted into the actual coordinates (X, Y) of the planar view. Finally, the vehicle planar contour information containing the vehicle's length, width, and shape is generated, which is represented as a set of polygonal edge points composed of a series of two-dimensional coordinate points. Camera calibration used Zhang Zhengyou's calibration method to obtain the camera intrinsic parameter matrix K and distortion coefficient k. A 7×5 checkerboard calibration plate with a side length of 30mm was selected. The camera intrinsic parameters were calculated by acquiring 20 sets of calibration images from different angles. Among them, focal length Pixels, principal point coordinates The perspective transformation matrix calculation defines the four corner points of the BEV image in the world coordinate system. Corresponding to the original image The mapping relationship, perspective transformation matrix M is solved using the least squares method, such that the transformed image satisfies the following conditions; The vertical alignment relationship of .

[0020] The ranging evaluation submodule calls the vehicle's planar contour information, sets multiple reference ranging points on the image width and height, calibrates the pixel coordinates corresponding to each ranging point, extracts pixel grayscale or color features for boundary recognition, calculates the pixel distance from the ranging point to the vehicle boundary, converts it into the spatial distance between the robot and the vehicle, and generates the corresponding distance evaluation value. Access vehicle planar contour information, such as a precise set of two-dimensional points ( Multiple reference ranging points are set in the width and height of the image. For example, in a 300×1200 pixel area, one point is set every 20×10 pixels, for a total of 500 points. The pixel coordinates of each ranging point are then calibrated. For example, (100, 50), extract the grayscale gradient or color features of a 5×5 pixel region, such as a gradient threshold of 30 and a color Euclidean distance threshold of 50, and perform boundary recognition, such as finding boundary pixels. The algorithm calculates the pixel distance from the ranging point to the vehicle boundary. For example, the pixel distance from point (100, 50) to boundary point (100, 70) is 20 pixels. This distance is then converted into the spatial distance between the robot and the vehicle. For example, using a scale of 0.5 mm per pixel, 20 pixels are converted to 10 mm. Combining the robot's position (1.5m, 0.8m) and the camera pose, the two-dimensional pixel distance is mapped to the three-dimensional spatial distance. For example, the actual spatial distance is 2.5 meters. Finally, the corresponding distance evaluation value is generated, which is a set of data containing the spatial coordinates of the ranging point and its corresponding spatial distance. To assess the distance between the robot and the vehicle, the module sets up five reference distance measurement points. Let the image width be W and the height be H, then the horizontal coordinates of the five distance measurement points are as follows: , , , , The vertical coordinates are all at the bottom of the image: These five ranging points are used to calculate the distance relationship between the robot and the vehicle, providing basic data for subsequent control algorithms.

[0021] The spatial information integration submodule calls the corresponding distance evaluation value, combines it with the vehicle planar contour information, performs coordinate mapping on the spatial position of the ranging point, merges adjacent or overlapping ranging results, removes abnormal deviation data, and outputs the vehicle spatial contour set. Call the corresponding distance evaluation value { }, and simultaneously call the vehicle's planar contour information to perform coordinate mapping on the spatial position of the ranging point, for example, by transferring the ranging point from the robot's local coordinate system ( Mapped to the world coordinate system This is achieved using the robot's GPS coordinates (34.2345, 108.9876, 500) and attitude data. It merges adjacent or overlapping distance measurement results. For example, a merging distance threshold of 20 mm is set. For overlapping points with a distance less than 20 mm and a relative distance difference of less than 5 mm, the weighted average of their spatial coordinates is calculated. The weight is determined by the confidence level (e.g., 0.9). Abnormal deviation data is removed. For example, an abnormal deviation threshold of 3 standard deviations from the median is set. If the deviation of the distance from the point to the vehicle contour (e.g., 100 mm) from the average distance exceeds this threshold, such as exceeding 15 mm (i.e., less than 85 mm or greater than 115 mm), it is removed. The final output is a set of vehicle spatial contour data, which is a set of integrated and denoised 3D point cloud data that accurately delineates the actual external shape of the vehicle.

[0022] Please see Figure 3 The image processing module includes: The denoising submodule is based on the vehicle spatial contour set, locates high-noise areas in the image, separates the background and target areas, performs pixel-level noise point removal, smooths local color and brightness changes, repairs disturbed contour segments, and generates a denoised image. Based on a vehicle spatial contour set, such as a point cloud containing (X, Y, Z) coordinates and RGB values, high-noise regions in the image are located. For example, the local 5x5 pixel gray-level variance is calculated, with a variance threshold of 25. If the variance is higher than 25 (e.g., 40), it is marked as a high-noise region. The background and target regions are separated, for example, by projecting the contour set onto the image to form a precise vehicle mask. Pixels within the mask are defined as target regions, and pixel-level noise point removal is performed. For example, the gray-level values ​​of the 8-neighborhood of a pixel are checked, and if the difference from the average value exceeds the noise threshold of 50 (0-255 range), the noise is removed. If the value is less than 10 pixels, it is replaced with the neighborhood average value, for example, pixel 200 is replaced with 100. Local color and brightness changes are smoothed, for example, by applying a 3x3 Gaussian filter kernel with a standard deviation of 0.8 and performing a weighted average on each pixel. Disturbed contour segments are repaired, for example, if the discontinuity length is less than 10 pixels, it is connected by linear interpolation or Bézier curve fitting. If the spur length is less than 5 pixels, it is repaired by morphological closing operation, for example, interpolation is performed at discontinuities of 8 pixels in length. Finally, a denoised image is generated, which is a clear 24-bit color image with background and target separation and smooth contours.

[0023] The edge enhancement submodule calls the denoised image, detects the vehicle boundary region, separates boundary pixels from non-boundary pixels, optimizes blurred edges, connects broken contour lines, balances boundary gray distribution, and generates an edge-enhanced image. The denoised image, such as a 24-bit RGB image of 1920×1080 pixels, is used to detect vehicle boundary regions. For example, the Canny edge detection operator is applied with a low threshold of 30 and a high threshold of 90 to extract preliminary edges. For instance, a pixel with a gradient magnitude of 120 at (500, 300) is identified as a strong edge. Boundary pixels are separated from non-boundary pixels to generate a binary image, where the vehicle outline is represented by a chain of white pixels. Blurred edges are then optimized by calculating the second derivative of the neighborhood of boundary pixels. If the absolute value is less than the blur threshold of 5, sub-blurring is performed. Pixel interpolation (such as bilinear interpolation) improves edge precision to 0.1 pixels, connects broken contour lines, for example, if the gap between breaks is less than 20 pixels, it connects them by straight line interpolation or curve fitting, for example, connecting the breakpoints (x1, y1) and (x2, y2), and balances the gray distribution of the boundary, for example, by performing local histogram equalization on the boundary pixels to adjust their gray value distribution to 0-255, so that the boundary contrast is consistent, and finally generates an edge-enhanced image, which is a binary image with clear, continuous vehicle contours and good contrast.

[0024] The boundary continuity assessment submodule calls the edge enhancement image, extracts the complete boundary pixel set, arranges the boundary points in spatial order, and uses the following formula: ; Calculate the distance difference between adjacent boundary points, mark missing or abrupt boundary segments, count the proportion of continuous and discontinuous segments, and generate a set of vehicle outline images. in, This represents the difference in spacing between adjacent boundary points. Representing the The boundary point and the first Euclidean distance between boundary points , Representing the first The horizontal and vertical coordinates of each boundary point , These represent the first neighbor used to calculate the average neighbor distance within the reference window. The horizontal and vertical coordinates of the nearest boundary points The number of neighboring boundary points selected for reference in the representative window. , Representing the first The horizontal and vertical coordinates of each boundary point; The edge enhancement image is invoked. For example, if the image is a binary image where the vehicle outline pixel value is 255 and the background pixel value is 0, the complete boundary pixel set is extracted. For example, by traversing the edge enhancement image, all connected pixels with a pixel value of 255 are extracted as boundary points, forming an ordered pixel chain. The boundary points are arranged in spatial order. For example, using a contour tracking algorithm, starting from a certain starting point, the pixel coordinates of all boundary points are tracked and recorded in a clockwise or counterclockwise direction according to the 8-neighborhood connectivity principle. , For example, for the left profile of a vehicle, the point sequence starts at (10, 500), followed by (10, 501), (11, 502), etc. The difference in distance between adjacent boundary points is calculated using the formula... ,in, Representing the The standardized difference value of the average distance between a boundary point and its neighboring points is used to measure the local continuity of the boundary. A larger value indicates that the boundary near that point is discontinuous or undergoes an abrupt change; if... A smaller value indicates a smooth and continuous boundary. Representing the boundary points ( , ) and the boundary points ( , The Euclidean distance between them is calculated as follows: This represents the actual pixel distance between adjacent boundary points. , Representing the current assessment The horizontal and vertical pixel coordinates of each boundary point are defined as follows: for example, the pixel coordinates are taken from the top left corner of the image as the origin, with the positive x-axis pointing to the right and the positive y-axis pointing downwards. , These represent the first neighbor used to calculate the average neighbor distance within the reference window. The horizontal and vertical pixel coordinates of the nearest boundary points, where the nearest points are around the point ( , In the boundary sequence, the M points before and after, The number of neighboring boundary points selected for reference in the representative window, for example, Setting it to 10 means calculating the average distance between the current point and its five nearest neighbors. , Representing the first The horizontal and vertical pixel coordinates of each boundary point, i.e., the next point in the boundary sequence from the current point. The numerator of this formula... Calculated the point ( ) to its The product of the sum of the Euclidean distances of the current point and the distance from the previous point to the current point reflects the overall spatial relationship between the current point and its local neighbors, and introduces local step size information. The denominator... This represents the product of the number of neighboring points and the Euclidean distance from the current point to the next point. This formula compares the average distance from the current point to its local neighborhood points with the distance from the neighboring points and records the offset abrupt change points, recording their positions. For example, if the offset value of a certain point segment is calculated to be 0.16 meters, it is marked as an abrupt change point, and its index in the path point sequence is recorded. The distance change between the path line and the vehicle contour is analyzed. For example, the Euclidean distance from each point on the path line to the nearest vehicle contour point is calculated, and its fluctuation range with the path is analyzed. For example, if the distance from the path point to the vehicle contour fluctuates between 10 centimeters and 12 centimeters, the fluctuation range is recorded. Finally, the path offset is generated. For example, the path offset is represented as structured data containing the average offset value of each point segment, the position of the abrupt change point, and its corresponding offset value.

[0025] The path adjustment interval determination submodule calls the path offset, combines the vehicle shape change characteristics, locates the start and end positions of the path segment that needs to be adjusted, sets the correction direction and correction magnitude of the adjusted segment, integrates the adjusted segment and the unadjusted segment, and generates a vehicle-around path planning table. The system retrieves path offset values, including average offset and abrupt change locations. It then combines these with vehicle morphology features, such as the curvature changes at the front and rear of the vehicle, to pinpoint the start and end points of the path segments requiring adjustment. For example, it sets an offset adjustment threshold of 0.08 meters. If the average offset of three consecutive point segments exceeds 0.08 meters, it marks them as the starting point for adjustment. For instance, three point segments starting from the 20th path point have an average offset of 0.12 meters. The system then sets the correction direction and magnitude for the adjusted segment. For example, if the vehicle deviates from the outer edge, the correction direction is to move inward along the normal direction, and the correction magnitude is the difference between the actual offset and the target safe distance of 0.05 meters. For example, if the actual offset is 0.15 meters, the correction magnitude is 0.10 meters. Finally, it integrates the adjusted and unadjusted segments, using spline interpolation to ensure path continuity and curvature smoothness at connections. The corrected path point list is then inserted into the original list and locally smoothed. This process ultimately generates a vehicle-around path planning table containing the 3D coordinates, tangent, normal, and corresponding speed and steering suggestions for all path points.

[0026] Please see Figure 6 The dynamic control module includes: The speed adjustment submodule extracts the robot's current position coordinates based on the vehicle path planning table, matches the reference points in the path point sequence from the current position, identifies the distance difference between the current position and the next reference point, determines the speed change range by combining the path curvature change, and generates a speed adjustment instruction set. Based on the vehicle path planning table, which includes path points and their expected speed and turning information, the robot's current position coordinates are extracted. For example, the three-dimensional position in the world coordinate system can be obtained in real time through the fusion of GPS and IMU. The system is accurate to the centimeter level, for example, (5.23m, 3.15m, 0.05m). It matches the current position to a reference point in the path point sequence. For example, it finds the path point P50 closest to the robot's current position as the current reference point, with a distance of 0.03 meters. It identifies the distance difference between the current position and the next reference point; for example, it calculates the Euclidean distance from the robot's current position to P51 (5.25m, 3.20m, 0.05m) to be approximately 0.0538 meters. It also combines this with changes in path curvature to determine the magnitude of speed changes, for example, by setting a curvature threshold of 0.1. If the curvature exceeds a threshold, the speed is reduced proportionally, for example, if the curvature is 0.2. The speed will decrease by 20%, if the curvature at the bend with a radius of 0.5 meters is 2.0. The speed is reduced from 0.5 m / s to 0.2 m / s, and a speed adjustment instruction set is generated, which includes the robot's target linear velocity, such as 0.2 m / s, and the adjustment trigger time.

[0027] The steering angle adjustment submodule calls the speed adjustment instruction set, combines the path point sequence and path offset, identifies the turning radius from the current path point to the next path point, determines the steering direction and angle range, corrects the steering angle under the influence of the offset, and generates the steering angle adjustment instruction set. The system invokes a speed adjustment instruction set, which includes the target turning angle and timestamp. It combines the path point sequence and path offset, such as the average offset value of each point segment and the location of abrupt change points, to identify the turning radius from the current path point to the next path point. For example, the radius is calculated using the current point, the next point, and the point after that, for example, 5.0 meters. The turning direction and angle magnitude are then determined, for example, based on the sign of the radius (positive for left turns). The smaller the radius, the larger the angle magnitude; for example, a 5.0-meter radius corresponds to a 10-degree turning angle magnitude. The system corrects the turning angle affected by the offset. For example, if the current path point has an outward offset of 0.05 meters, a correction angle proportional to the offset is added to the turning angle. For example, every 0.01-meter offset increases by 0.5 degrees, so 0.05 meters increases by 2.5 degrees, making the original 10-degree turning angle 12.5 degrees. Finally, a turning angle adjustment instruction set is generated, including the robot's target turning angle, for example, 12.5 degrees, and the corresponding trigger time.

[0028] The execution time synchronization submodule calls the steering angle adjustment instruction set, extracts the timestamps of the speed adjustment and steering angle adjustment instructions, calculates the time difference between the two, adjusts the execution order of the instructions, synchronizes the corresponding execution start and end times, and combines the synchronized speed and steering instructions to obtain the dynamic control signal set of the car wash robot. Call the steering angle adjustment command set and the speed adjustment command set. For example, both contain target values ​​and timestamps. Extract the timestamps of the speed adjustment and steering angle adjustment commands, such as the timestamp of speed command V1. The timestamp for the turning instruction A1 is 1678886400.123 seconds. The time difference between the two is 1678886400.125 seconds. For example, 0.002 seconds, adjusting the order of instruction execution, for example... If the timeout exceeds the synchronization tolerance threshold of 0.001 seconds, the instruction with the later timestamp will be delayed to synchronize with the earlier instruction. For example, if... Earlier If the value exceeds 2 milliseconds, the A1 instruction will be delayed by 2 milliseconds to synchronize the start and end times of the corresponding instruction. For example, this ensures that all instructions start execution within the same millisecond. The synchronized speed and steering instructions are combined. For example, the synchronized linear speed of 0.5 m / s and the steering angle of 15 degrees are packaged into a unified control signal data packet. Finally, the dynamic control signal set of the car wash robot is obtained. It is a continuous, time-aligned control instruction stream that contains all the real-time motion parameters required for robot drive.

[0029] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A multi-camera coordinated car washing robot path planning system, characterized in that, The system includes: The vision acquisition module uses multiple sets of cameras distributed on the top and sides of the robot to simultaneously acquire multi-angle images of the vehicle, record timestamps and corresponding poses, perform pixel-level distortion correction and viewpoint correction, identify the vehicle's planar contour information, and generate a set of vehicle spatial contours. The image processing module performs pixel-level noise suppression, differential scale detail adjustment, edge enhancement and continuity restoration, and merges contour refinement and connected components based on the vehicle spatial contour set to generate a vehicle contour image set. The contour extraction module calls the vehicle contour image set, detects the outer boundary pixels of the vehicle, performs fine localization of boundary points, samples and sorts boundary points, removes isolated noise points, identifies the curvature and normal of boundary points, segments straight lines and curves, densely samples high curvature areas, and generates a vehicle contour boundary set. Based on the vehicle contour boundary set, the path planning module sets the robot's working distance and lateral margin, identifies the offset curve on the outside of the vehicle body, samples path points along the offset curve by arc length, adjusts the path point spacing to match the robot's motion resolution, and removes points that overlap with the vehicle contour to obtain the vehicle-around path planning table. The dynamic control module calls the vehicle path planning table, estimates the path index corresponding to the current pose, identifies lateral and heading errors, combines the velocity profile, calculates the desired speed and heading by interpolation, applies acceleration / deceleration and steering angular velocity limits, outputs servo and drive control commands, monitors execution errors, marks trajectory deviations, and obtains the dynamic control signal set of the car wash robot. The dynamic control signal set of the car wash robot includes speed adjustment commands, steering angle adjustment commands, and execution time synchronization information; The dynamic control module includes: Based on the vehicle path planning table, the speed adjustment submodule extracts the robot's current position coordinates, matches the reference points in the path point sequence from the current position, identifies the distance difference between the current position and the next reference point, determines the speed change range by combining the path curvature change, and generates a speed adjustment instruction set. The steering angle adjustment submodule calls the speed adjustment instruction set, combines the path point sequence and path offset, identifies the turning radius from the current path point to the next path point, determines the steering direction and angle range, corrects the steering angle under the influence of the offset, and generates the steering angle adjustment instruction set. The execution time synchronization submodule calls the steering angle adjustment instruction set, extracts the timestamps of the speed adjustment and steering angle adjustment instructions, calculates the time difference between the two, adjusts the execution order of the instructions, synchronizes the corresponding execution start and end times, and combines the synchronized speed and steering instructions to obtain the dynamic control signal set of the car wash robot.

2. The multi-camera coordinated car washing robot path planning system of claim 1, wherein, The vehicle spatial contour set includes vehicle planar contour information, boundary point distribution features, and corresponding distance evaluation values. The vehicle contour image set includes denoised images, edge-enhanced images, and boundary continuity evaluation values. The vehicle contour boundary set includes boundary point sequences, boundary distribution features, and boundary continuity indices. The vehicle-around path planning table includes path point sequences, path offsets, and path adjustment intervals.

3. The multi-camera coordinated car washing robot path planning system of claim 1, wherein, The visual acquisition module includes: The multi-view acquisition submodule uses multiple cameras on the top and sides of the robot to perform multi-camera synchronous startup and exposure control, acquire multi-frame images of the vehicle from different angles, record the timestamp of each frame and camera pose parameters, perform pixel distortion correction and image registration, and use inverse perspective transformation to convert the image into a planar view to generate vehicle planar contour information. The ranging evaluation submodule calls the vehicle's planar contour information, sets multiple reference ranging points on the image width and height, calibrates the pixel coordinates corresponding to each ranging point, extracts pixel grayscale or color features for boundary recognition, calculates the pixel distance from the ranging point to the vehicle boundary, converts it into the spatial distance between the robot and the vehicle, and generates the corresponding distance evaluation value. The spatial information integration submodule calls the corresponding distance evaluation value, combines it with the vehicle planar contour information, performs coordinate mapping on the spatial position of the ranging point, merges adjacent or overlapping ranging results, removes abnormal deviation data, and outputs the vehicle spatial contour set.

4. The multi-camera coordinated car washing robot path planning system of claim 3, wherein, The image processing module includes: The denoising submodule, based on the vehicle spatial contour set, locates high-noise areas in the image, separates the background and target areas, performs pixel-level noise point removal, smooths local color and brightness changes, repairs disturbed contour segments, and generates a denoised image. The edge enhancement submodule calls the denoised image, detects the vehicle boundary region, separates boundary pixels from non-boundary pixels, optimizes blurred edges, connects broken contour lines, balances boundary grayscale distribution, and generates an edge-enhanced image. The boundary continuity assessment submodule calls the edge enhancement image, extracts the complete boundary pixel set, arranges the boundary points in spatial order, calculates the distance difference between adjacent boundary points, marks missing or abrupt boundary segments, calculates the proportion of continuous and discontinuous segments, and generates a vehicle contour image set.

5. The multi-camera collaborative car wash robot path planning system according to claim 4, characterized in that, The contour extraction module includes: The boundary point recognition submodule locates the area where the vehicle's outer boundary is located based on the vehicle contour image set, separates the boundary area from the non-boundary area, scans the pixels of the boundary area point by point, records the pixel coordinates that conform to the outer boundary features, sorts the boundary points in spatial order, and obtains the boundary point distribution features. The boundary distribution analysis submodule calls the boundary point distribution characteristics, calculates the distribution density of boundary points in the horizontal and vertical directions, counts the range of variation in the distance between boundary points, detects the location of dense and sparse areas, analyzes the changes in the direction of boundary points, and obtains boundary distribution characteristic data. The contour integrity assessment submodule calls the boundary distribution feature data, reconstructs the boundary point connection to form a closed contour, detects the length and position of contour gaps, calculates the ratio of continuous segments to broken segments, measures the smoothness of contour lines, marks missing boundary segments, and generates a vehicle contour boundary set.

6. The multi-camera collaborative car wash robot path planning system according to claim 5, characterized in that, The path planning module includes: The path point calculation submodule extracts the spatial coordinates of boundary points based on the vehicle outline boundary set, generates path reference points along the vehicle outline at fixed intervals, identifies the tangential and normal directions of the path reference points, corrects points that do not match the vehicle outline, connects the path reference points to form a continuous path line, and obtains the path point distribution information. The path offset analysis submodule calls the path point distribution information, analyzes the displacement difference between adjacent path points in the horizontal and vertical directions, counts the average offset of continuous point segments, calculates the average offset value of continuous point segments, detects offset abrupt change points and records their positions, analyzes the distance change between the path line and the vehicle outline, and generates the path offset. The path adjustment interval determination submodule calls the path offset, combines the vehicle shape change characteristics, locates the start and end positions of the path segment that needs to be adjusted, sets the correction direction and correction magnitude of the adjusted segment, integrates the adjusted segment and the unadjusted segment, and generates a vehicle-around path planning table.

Images