Agv system for vehicle chassis corrosion assessment

Abstract

An AGV system for vehicle chassis corrosion evaluation, through a vehicle bottom real-time positioning module, particle filtering and image matching processing are respectively carried out according to point cloud information collected by a plane laser scanner and image information collected by a camera, and an abnormal value filtering is carried out on the positioning result, so that a vehicle bottom laser and image fusion positioning result is obtained; through a corrosion evaluation module, part complexity evaluation, adaptive part region segmentation and corrosion grade evaluation processing are carried out according to the vehicle bottom positioning information and real-time image information collected by the camera, so that a vehicle bottom part corrosion grade evaluation result is obtained; through a corrosion review module, strong edge suppression, local entropy calculation and neural network judgment corrosion processing are carried out according to part local corrosion image information, so that a part corrosion review result is obtained, the detected corrosion area can be reviewed, false corrosion areas such as red paint parts are screened out, and the corrosion grade evaluation precision is improved.

Description

Technical Field

[0001] This invention relates to a technology in the field of vehicle maintenance, specifically an AGV system for assessing corrosion of vehicle chassis. Background Technology

[0002] Existing chassis corrosion assessment solutions primarily utilize lifts, fixed ground-based inspection systems, and mobile inspection trolleys. Among these, mobile inspection trolleys, capable of free movement under the vehicle and capturing chassis images via cameras, offer good portability and high inspection efficiency. However, current mobile inspection trolley solutions typically transmit chassis images to maintenance personnel in real-time or stitch together fragmented real-time images to form a complete chassis image. They lack the ability to autonomously detect and assess corrosion areas from the acquired images, resulting in low levels of automation and intelligence. Summary of the Invention

[0003] This invention addresses the shortcomings of existing testing devices, such as their immobility and low automation, by proposing an AGV system for vehicle chassis corrosion assessment. This system can autonomously traverse the vehicle's undercarriage space, detect corrosion areas, and assess corrosion levels without requiring pre-prepared chassis component datasets. It accurately assesses corrosion levels based on component corrosion area coverage, ultimately generating a valuable chassis corrosion assessment report and saving images of key areas for convenient comprehensive chassis condition evaluation by maintenance personnel. Furthermore, by integrating convolutional neural networks and image local entropy values ​​into a chassis corrosion area verification method, the system can verify detected corrosion areas, filtering out false corrosion areas such as red-painted parts, thus improving the accuracy of corrosion level assessment.

[0004] This invention is achieved through the following technical solution:

[0005] This invention relates to an AGV system for assessing corrosion of vehicle chassis, comprising: a real-time undercarriage positioning module, a corrosion assessment module, and a corrosion verification module. The real-time undercarriage positioning module performs particle filtering and image matching processing on point cloud information acquired by a planar laser scanner and image information acquired by a camera, and performs outlier filtering on the positioning results to obtain a fusion positioning result of the laser and image data. The corrosion assessment module performs part complexity assessment, adaptive part region segmentation, and corrosion level assessment processing on the undercarriage positioning information and real-time image information acquired by the camera to obtain a corrosion level assessment result for the undercarriage parts. The corrosion verification module performs strong edge suppression, local entropy calculation, and neural network-based corrosion judgment processing on the local corrosion image information of the parts to obtain a corrosion verification result for the parts.

[0006] This invention relates to a method for assessing corrosion of vehicle chassis based on the above-mentioned AGV system, comprising the following steps:

[0007] Step 1: Correct the AGV's positioning under the vehicle based on the point cloud and image information collected by the AGV;

[0008] Step 2: Perform image preprocessing on the chassis images captured by the camera mounted on the AGV;

[0009] Step 3: Evaluate the complexity of parts in the chassis images;

[0010] Step 4: Based on the part complexity assessment results, perform adaptive part region segmentation on the chassis image;

[0011] Step 5: Detect corrosion areas in the chassis image using the color space threshold filtering method and assess the corrosion level;

[0012] Step 6: Perform strong edge suppression on the eroded areas of the chassis image and then perform local entropy erosion verification;

[0013] Step 7: Perform neural network erosion verification on the eroded areas of the chassis image and fuse the erosion verification results;

[0014] Step 8: Add positioning tags to the corrosion assessment and verification results based on the AGV's real-time positioning information.

[0015] Technical effect

[0016] This invention utilizes an adaptive part region segmentation technique based on part complexity assessment and a corrosion verification technique based on a fusion of local entropy and neural networks with strong edge suppression. This technique enables adaptive segmentation of parts of different sizes and types based on part complexity assessment, and calculates and evaluates the corrosion level separately without the need to prepare a part dataset in advance. The corrosion region is verified by a fusion of local entropy and neural networks, which eliminates false detections of non-corroded parts that can occur with the color space threshold filtering method. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the present invention;

[0018] Figure 2 A schematic diagram of the appearance corrosion evaluation criteria defined in document GMW-15357;

[0019] Figure 3 This is a schematic diagram of a simulation scene;

[0020] Figure 4 , 5 The red dots in the constructed laser map represent real-time point cloud data.

[0021] Figure 6 , 7 A schematic diagram of the AGV's traversal path for corrosion assessment;

[0022] Figure 8A schematic diagram of the chassis image after adaptive histogram equalization processing;

[0023] Figure 9 This is a schematic diagram of strong edge detection;

[0024] Figure 10 This is a schematic diagram of the local entropy of the part's contour.

[0025] Figure 11 This is a schematic diagram of weak edge detection;

[0026] Figure 12 This is a schematic diagram of strong edge suppression;

[0027] Figure 13 A schematic diagram of the adaptive Filsenzwab oversegmentation results;

[0028] Figure 14 This is a schematic diagram of the adaptive region adjacency graph merging result;

[0029] Figure 15 This is a schematic diagram of the part region segmentation results after morphological closed region detection;

[0030] Figure 16 , 17 A schematic diagram showing the segmentation of etched and non-etched regions using the HSV color space threshold.

[0031] Figure 18 A schematic diagram showing the corrosion assessment results for different areas of the component;

[0032] Figure 19 A schematic diagram showing the segmentation of the corrosion sub-regions of the part;

[0033] Figure 20 This is a schematic diagram of the local entropy corrosion verification of the corrosion sub-region of the part;

[0034] Figure 21 A schematic diagram of neural network corrosion verification of the corrosion sub-region of the part;

[0035] Figure 22 This is a schematic diagram illustrating the fusion logic of local entropy and neural network erosion verification results. Detailed Implementation

[0036] like Figure 1As shown in the figure, this embodiment relates to an AGV system for vehicle chassis corrosion assessment, including: a real-time vehicle undercarriage positioning module, a corrosion assessment module, and a corrosion verification module. The real-time vehicle undercarriage positioning module performs particle filtering and image matching processing on point cloud information collected by a planar laser scanner and image information collected by a camera, and performs outlier filtering on the positioning results to obtain a vehicle undercarriage laser and image fusion positioning result. The corrosion assessment module performs part complexity assessment, adaptive part region segmentation, and corrosion level assessment processing on real-time image information collected by the camera to obtain a corrosion level assessment result for the vehicle undercarriage parts. The corrosion verification module performs strong edge suppression, local entropy calculation, and neural network-based corrosion judgment processing on the vehicle undercarriage positioning information and local corrosion image information of the parts to obtain a corrosion verification result for the parts and marks the corrosion areas of the parts.

[0037] The real-time vehicle undercarriage positioning module includes: a laser map construction unit, a chassis image stitching unit, a map calibration unit, and a vehicle undercarriage positioning correction unit. Specifically: the laser map construction unit performs particle filtering based on information from a planar laser scanner to obtain a laser map of the vehicle undercarriage; the image map construction unit performs image feature matching and stitching based on image information acquired by a camera to obtain a chassis image map; the map calibration unit performs coordinate system transformation based on the relative positions of the vehicle tires on the laser map and the image map to obtain a map transformation matrix; and the vehicle undercarriage positioning correction unit performs outlier filtering based on the map transformation matrix and the AGV's real-time positioning information to obtain a corrected vehicle undercarriage positioning result.

[0038] The corrosion assessment module includes: an image preprocessing unit, a part complexity assessment unit, a part region segmentation unit, a corrosion region segmentation unit, and a corrosion level assessment unit. Specifically: the image preprocessing unit performs adaptive histogram equalization, downsampling, and Gaussian blurring on the image information acquired by the camera to obtain a preprocessed image result; the part complexity assessment unit performs strong edge detection, local entropy calculation, and part complexity calculation on the preprocessed image information to obtain a part complexity result; the part region segmentation unit performs adaptive Filsen-Zwab segmentation, adaptive region adjacency graph merging, and morphological closure detection on the preprocessed image and part complexity information to obtain a part region segmentation result; the corrosion region segmentation unit performs HSV color space filtering on the preprocessed image information to obtain a corrosion region segmentation result; and the corrosion level assessment unit performs corrosion coverage calculation and local corrosion region segmentation on the part region and corrosion region segmentation information to obtain a corrosion assessment level and a local corrosion image result.

[0039] The corrosion verification module includes: a local entropy corrosion verification unit, a neural network corrosion verification unit, and a corrosion region marking unit. The local entropy corrosion verification unit performs strong edge suppression and corrosion entropy calculation based on local corrosion image information to obtain a local entropy corrosion verification result. The neural network corrosion verification unit performs neural network binary classification based on local corrosion image information to obtain a neural network corrosion verification result. The corrosion region marking unit fuses the verification results and performs image marking processing based on the local entropy, neural network corrosion verification information, and corrected vehicle undercarriage positioning information to obtain a marked corrosion region result.

[0040] like Figure 1 As shown, this embodiment relates to a method for assessing corrosion of the vehicle chassis of the aforementioned AGV system, including the following steps:

[0041] Step 1: Correct the AGV's positioning under the vehicle based on the point cloud and image information collected by the AGV. This specifically includes:

[0042] 1.1) Laser Map Construction: Place the AGV under the vehicle to be inspected, allowing the AGV to randomly explore and construct a laser map under the vehicle, such as... Figure 3 As shown in the diagram. In this step, because the planar laser scanner of the chassis corrosion assessment AGV is installed at a low height, and the distance between the vehicle tires is much smaller than the 12-meter detection range of the laser scanner, the AGV can scan the tire contour point cloud in real time. By randomly selecting target points in the space under the vehicle, the AGV can autonomously explore the underside of the vehicle until a complete laser map of the underside is constructed, as shown in the diagram. Figures 4-7 As shown, when there are no other obstructions under the vehicle, the laser map under the vehicle is represented by four rectangles representing the tires.

[0043] In this embodiment, a laser map coordinate system for the vehicle's underside is constructed with the vehicle's center as the origin, the vehicle's front direction as the positive y-axis, and the vehicle's rightward direction as the positive x-axis. The transpose of the coordinate matrix in this laser coordinate system, formed by the centers of the four tires (left front tire, right front tire, left rear tire, and right rear tire), is:

[0044] 1.2) Image Map Construction: The AGV simultaneously acquires chassis images in real time via a USB camera with a vertically upward viewing angle. SURF feature points are detected in the real-time images, and FLANN feature matching is performed on the SURF feature points of adjacent frames to calculate the homography matrix between adjacent frames. Based on the homography matrix H, adjacent frame images are mapped and transformed, and overlapping parts are stitched together to finally form a complete chassis image map.

[0045] In this embodiment, the chassis image map is a pixel coordinate system, with the origin at the upper left corner of the image. The horizontal direction to the right is the positive u-axis, and the vertical direction downwards is the positive v-axis. Therefore, the transpose of the coordinate matrix in the pixel coordinate system formed by the centers of the four tires (left front tire, right front tire, left rear tire, and right rear tire) is:

[0046] 1.3) Map Delineation: Based on the coordinate matrices of the centers of the four tires (left front tire, right front tire, left rear tire, and right rear tire) in the laser and pixel coordinate systems, calculate the coordinate transformation matrix between the laser map and the image map. Map mapping complete.

[0047] In this embodiment, for any coordinate under the laser map Its coordinates in the image map can be obtained through coordinate transformation. Similarly, for any coordinates under an image map Coordinate transformation matrix can be used Find its coordinates on the laser map.

[0048] 1.4) Vehicle undercarriage positioning correction: The coordinate system transformation and outlier filtering are performed on the positioning results of the image map and laser map to correct the vehicle undercarriage positioning results.

[0049] Because laser map localization is prone to pose estimation flipping issues, in this embodiment, the laser map localization results... Image map positioning results obtained through coordinate system transformation If x is detected in three consecutive frames s and x p If the sign is reversed, the laser map is flipped along the y-axis. Similarly, if y is detected in three consecutive frames... s and y p If the sign is reversed, the laser map will be flipped along the x-axis.

[0050] Image map localization methods are prone to mismatch issues. In this embodiment, the image map localization results... The laser map positioning results obtained after coordinate system transformation if |u p -u s |>0.1*u max or |v p -v s |>0.1*v max If the absolute value of the difference between the positioning results of the image map and the laser map exceeds 10% of the total size of the image map, the positioning result of the image map is considered an outlier, and the corrected positioning result is taken from the laser map P′.s .

[0051] In this embodiment, for the image map positioning results The laser map positioning results obtained after coordinate system transformation if |u p -u s |≤0.1*u max or |v p -v s |≤0.1*v max If the absolute value of the difference between the positioning results from the image map and the laser map does not exceed 10% of the total size of the image map, then both positioning results are considered to be normal, and the corrected positioning result is the average of the two.

[0052] Step 2: Perform image preprocessing on the chassis images captured by the camera mounted on the AGV, specifically including:

[0053] 2.1) Adaptive Histogram Equalization: Adaptive histogram equalization is performed on the raw chassis image acquired by the camera, such as... Figure 8 As shown. This operation can effectively brighten insufficiently lit areas while preserving part details.

[0054] In this embodiment, the equalization threshold is 0.1.

[0055] 2.2) Downsampling: Downsampling is performed on the RGB three channels of the chassis image. This operation can blur high-frequency details of the parts and highlight the contour curves of the parts.

[0056] In this embodiment, the selected downsampling ratio is:

[0057] 2.3) Gaussian Blur: Apply a Gaussian blur to the chassis image. This operation can blur high-frequency details of parts and highlight the contour curves of the parts.

[0058] In this embodiment, a 5×5 convolution kernel is selected, with a standard deviation σ = 0.8.

[0059] Step 3: Evaluate the component complexity of the chassis image, specifically including:

[0060] 3.1) Strong Edge Detection: Perform Canny strong edge detection on the chassis image, such as... Figure 9 As shown. By setting appropriate Gaussian blur standard deviation and link threshold, this operation can detect strong edges of parts, i.e., part contours, and remove weak edges below the link threshold.

[0061] In this embodiment, the standard deviation of the Gaussian blur is selected as σ = 3, the minimum link threshold is min_val = 0.10, and the maximum link threshold is max_val = 0.20.

[0062] 3.2) Part Complexity Assessment: Based on the strong edge detection results in the image, the part complexity is assessed. Generally speaking, the more strong edges in the image, i.e., the more part outlines, the higher the part complexity in the image. Simultaneously, the higher the local entropy value of the image, the more uneven the distribution of parts, and the higher the part complexity.

[0063] In this embodiment, based on the number of strong edge pixels n e and the total number of pixels in the image n p Calculate the ratio of strong edges in an image. The range of values ​​is 0 ≤ R e ≤1.

[0064] In this embodiment, the local entropy expression is used within a 5×5 rectangular moving window. Calculate the local entropy for the strong edge detection results, such as Figure 10 As shown. Since the strong edge detection result is a binary image, the local entropy expression can be simplified to H. e = -(p0log2p0+(1-p0)log2(1-p0)). Differentiating this expression yields... When the probability p0 of a pixel intensity being 0 is taken When the number of pixels with intensities of 1 and 0 in the moving window is the same, the local entropy H is... e It can take the maximum value of 1. Meanwhile, when p0 is 0 or 1, the local entropy H... e Take the minimum value of 0. Take the average value of the local entropy of the image. The range of values ​​is also...

[0065] In this embodiment, part complexity evaluation parameters are proposed. The range of values ​​is 0 ≤ C e ≤1. C e The higher the value, the more complex the parts in the image.

[0066] 3.3) Adaptive parameter calculation: Based on the calculated part complexity evaluation parameter C e The adaptive computation Filsen-Zwab segmentation algorithm is applicable to the scale parameter (scale), and the region adjacency graph algorithm is applicable to the merging threshold (thresh).

[0067] In this embodiment, for images with high part complexity, a low-scale parameter is selected. For images with width w and height h, the applicable scale parameter for the Filsenzwab segmentation algorithm is proposed and calculated. Where k is the amplification factor, and 5 is selected in this embodiment.

[0068] In this embodiment, for images with high part complexity, a low merging threshold is selected, and the appropriate merging threshold for the region adjacency graph algorithm is proposed and calculated.

[0069] Step 4: Based on the part complexity assessment results, perform adaptive part region segmentation on the chassis image, specifically including:

[0070] 4.1) Adaptive Filsenzwab segmentation: based on the scale parameters obtained through adaptive calculation. Adaptive Felsenberg segmentation is performed on the preprocessed chassis image to obtain oversegmentation results of the part regions, such as... Figure 13 As shown.

[0071] In this embodiment, the minimum segmentation region size min_size = scale is selected.

[0072] 4.2) Adaptive Region Adjacency Graph Merging: Based on the adaptively calculated merging threshold... The oversegmentation results of the part regions are then subjected to adaptive region adjacency graph merging based on color similarity to obtain preliminary part region segmentation results, such as... Figure 14 As shown.

[0073] 4.3) Morphological Closure Detection: For part regions with similar colors but not spatially adjacent areas, morphological closure detection is performed to segment the part region into closed region instances, obtaining the final part region segmentation result, such as... Figure 15 As shown.

[0074] Step 5: Detect the corroded areas in the chassis image using a color space threshold filtering method and evaluate the corrosion level. This specifically includes:

[0075] 5.1) HSV color space filtering: Converts the image from the RGB color space to the HSV color space, based on a preset threshold filtering.

[0076] In this embodiment, for the HSV image, pixels with H channel values ​​falling within the [0,15] and [170,180] intervals, i.e., red pixels, are segmented and used as erosion region masks, such as... Figure 16 , 17 As shown.

[0077] 5.2) Component Corrosion Level Evaluation: Perform bitwise AND operations between different component areas and the corrosion area mask to obtain the number of non-zero pixels e. i Based on the total number of pixels n in the part area i Calculate the overlap rate like Figure 18 As shown.

[0078] In this embodiment, the corrosion level is assessed for different parts based on the corrosion coverage rate as defined in General Motors' GMW-15357 document. For example, for corrosion coverage rate K... i =8% of the parts were assessed as having level 7 corrosion, which is considered slight corrosion.

[0079] 5.3) Part Corrosion Sub-region Segmentation: Corrosion areas with corrosion coverage exceeding a set threshold are segmented using a minimum rectangular bounding box to facilitate verification by the corrosion review module. Figure 19 As shown.

[0080] In this embodiment, the corrosion coverage verification threshold is set to K. i ≥5%. Figure 19 The corrosion coverage K of the corroded sub-region of the part shown i The corrosion rate was 36.85%, which was assessed as Level 6 corrosion, belonging to moderate corrosion.

[0081] Step 6: Perform strong edge suppression on the eroded areas of the chassis image and conduct local entropy erosion verification, specifically including:

[0082] 6.1) Strong Edge Suppression: Canny edge detection with different parameters is performed on the corroded sub-regions of the part, and the difference is used to obtain the corrosion edge map. With relatively low Gaussian blur standard deviation and link threshold, Canny edge detection can detect all edges, including strong edges of the part outline and weak edges of corrosion details. Figure 11 As shown. When the Gaussian blur standard deviation and link threshold are relatively high, Canny edge detection only detects strong edges on the part contour, such as... Figure 9 As shown.

[0083] In this embodiment, the strong edges of the part contour are dilated using a 3×3 convolution kernel, and the difference is calculated with the edge detection result obtained under the low link threshold condition to preserve the edge map of the part's corrosion details, such as... Figure 12 As shown.

[0084] In this embodiment, the Gaussian blur standard deviation σ = 1 is used for weak edge detection, the minimum link threshold min_val = 0.05, and the maximum link threshold max_val = 0.10. The Gaussian blur standard deviation σ = 3 is used for strong edge detection, the minimum link threshold min_val = 0.10, and the maximum link threshold max_val = 0.20.

[0085] 6.2) Local Entropy Corrosion Verification: The average local entropy is calculated on the corrosion detail edge map of the part. Corrosion sub-regions with an average local entropy greater than a set threshold are verified as having undergone corrosion. Figure 20 As shown.

[0086] In this embodiment, the local entropy expression H e =-(p0log2p0+(1-p0)log2(1-p0)), the corrosion threshold is selected as Figure 20 The average local entropy of the corroded sub-region of the part shown is 0.38. The local entropy corrosion verification result is positive, indicating that corrosion has occurred.

[0087] Step 7: Perform neural network erosion verification on the eroded areas of the chassis image, and fuse the erosion verification results, specifically including:

[0088] 7.1) Neural Network Binary Classification Verification: A pre-trained neural network binary classification model is used to perform a binary classification verification of corrosion / non-corrosion in the corrosion sub-regions of the part, such as... Figure 21 As shown.

[0089] In this embodiment, a VGG16 convolutional neural network is trained based on a metal corrosion dataset, with the sigmoid function as the activation function and the binary cross-entropy as the loss function, to generate a pre-trained binary classification model.

[0090] 7.2) Fusion of Corrosion Verification Results: A bitwise OR operation is performed on the local entropy and neural network corrosion verification results. If either result indicates the presence of corrosion, the final determination is that corrosion exists. The specific corrosion judgment logic is as follows: Figure 22 As shown.

[0091] Step 8: Based on the AGV's real-time positioning information, add positioning tags to the corrosion assessment and verification results, specifically including:

[0092] 8.1) Corrosion area marking: The corrosion sub-regions of the parts, corrosion assessment results, and corrosion verification results are combined with the corrected undercarriage positioning coordinates and marked on the image map as the final corrosion assessment report.

[0093] 8.2) Image saving of corrosion areas: Images of parts identified as severely corroded are saved locally on the AGV for maintenance personnel to refer to.

[0094] Compared with existing technologies, this method adaptively segments parts of different sizes and types based on part complexity assessment and calculates and evaluates corrosion levels separately, without the need to prepare part datasets in advance; it performs local entropy and neural network fusion verification on the corroded areas to screen out false detections of non-corroded parts that exist in the color space threshold filtering method; and it is based on joint localization of laser maps and image maps, which has good versatility for vehicles of different sizes and types and is not limited by the site.

[0095] The above-described specific implementations can be partially adjusted by those skilled in the art in different ways without departing from the principles and purpose of the present invention. The scope of protection of the present invention is defined by the claims and is not limited to the above-described specific implementations. All implementation schemes within the scope of the claims are bound by the present invention.

Claims

1. An AGV system for vehicle chassis corrosion assessment, characterized in that, The application relates to a vehicle chassis real-time positioning and corrosion evaluation system. The part complexity evaluation comprises: a) strong edge detection: Canny strong edge detection is carried out on the chassis image; The adaptive part region segmentation comprises: b) Part complexity evaluation: According to the strong edge detection result of the image, the part complexity is evaluated, that is, the strong edge pixel point number and the total pixel point number of the image are calculated , the strong edge ratio of the image is calculated , the value range is , and the part complexity evaluation parameter ; c) adaptive parameter calculation: based on the calculated part complexity assessment parameter , the adaptive calculation of the Felzenszwalb segmentation algorithm applicable scale parameter and the region adjacency graph algorithm applicable merging threshold thresh; The Felzenszwalb segmentation algorithm is applicable to a scale parameter scale = wherein w is the width of the image, h is the height of the image, and k is a magnification factor. The merging threshold thresh = 0.5 ; 3) morphological closure detection: morphological closure detection is carried out on part regions which are similar in color but not adjacent in space, the part regions are segmented into closed region instances, and finally, the part region segmentation result is obtained. 1) Adaptive Felzenszwalb segmentation: according to the adaptive calculation scale parameter scale = The pre-processed chassis image is subjected to adaptive Felzenszwalb segmentation to obtain a part region over-segmentation result; 2) adaptive region adjacency graph merging: according to the adaptive calculated merging threshold thresh = perform adaptive region adjacency graph merging based on color similarity on the part region over-segmentation result to obtain a preliminary part region segmentation result; The vehicle chassis real-time positioning module comprises a laser map construction unit, a chassis image splicing unit, a map calibration unit and a vehicle chassis positioning correction unit, wherein: the laser map construction unit carries out particle filtering processing according to the plane laser scanner information, and obtains the vehicle chassis laser map result; the image map construction unit carries out image feature matching and splicing processing according to the image information collected by the camera, and obtains the chassis image map result; the map calibration unit carries out coordinate system conversion processing according to the relative position information of the vehicle tire in the laser map and the image map, and obtains the map conversion matrix result; and the vehicle chassis positioning correction unit carries out abnormal value filtering processing according to the map conversion matrix and the AGV real-time positioning information, and obtains the corrected vehicle chassis positioning result.

2. The AGV system for vehicle underbody corrosion assessment of claim 1, wherein, The corrosion evaluation module comprises an image preprocessing unit, a part complexity evaluation unit, a part region segmentation unit, a corrosion region segmentation unit and a corrosion grade evaluation unit, wherein: the image preprocessing unit carries out adaptive histogram equalization, down-sampling and Gaussian blur processing on the image information collected by the camera, and obtains the preprocessed image result; the part complexity evaluation unit carries out strong edge detection, local entropy calculation and part complexity calculation processing on the preprocessed image information, and obtains the part complexity result; the part region segmentation unit carries out adaptive Felzenszwalb segmentation, adaptive region adjacency graph merging and morphological closure detection processing on the preprocessed image and the part complexity information, and obtains the part region segmentation result; the corrosion region segmentation unit carries out HSV color space filtering processing on the preprocessed image information, and obtains the corrosion region segmentation result; and the corrosion grade evaluation unit carries out corrosion coverage rate calculation and local corrosion region segmentation processing on the part region and the corrosion region segmentation information, and obtains the corrosion evaluation grade and the local corrosion image result.

3. The AGV system for vehicle underbody corrosion assessment of claim 1, wherein, ​ 4. The AGV system for vehicle underbody corrosion assessment of claim 1, wherein, The corrosion review module comprises a local entropy corrosion review unit, a neural network corrosion review unit and a corrosion area marking unit, wherein: the local entropy corrosion review unit performs strong edge suppression and corrosion entropy calculation processing according to local corrosion image information to obtain a local entropy corrosion review result; the neural network corrosion review unit performs neural network binary classification processing according to the local corrosion image information to obtain a neural network corrosion review result; and the corrosion area marking unit performs review result fusion and image marking processing according to the local entropy and neural network corrosion review information and the corrected vehicle bottom positioning information to obtain a marked corrosion area result.

5. A method of assessing the corrosion of a vehicle chassis according to the system of any one of claims 1 to 4, characterized in that, The method comprises the following steps: Step 1: correcting the positioning of the AGV on the vehicle bottom according to the point cloud and image information collected by the AGV; Step 2: performing image preprocessing on the chassis image collected by the camera carried by the AGV; Step 3: performing part complexity evaluation on the chassis image; Step 4: performing adaptive part region segmentation on the chassis image according to the part complexity evaluation result; Step 5: detecting the corrosion area of the chassis image based on the color space threshold filtering method and evaluating the corrosion level; Step 6: performing strong edge suppression on the corrosion area of the chassis image and performing local entropy corrosion review; Step 7: performing neural network corrosion review on the corrosion area of the chassis image and fusing the corrosion review results; Step 8: adding a positioning label to the corrosion evaluation and review results according to the real-time positioning information of the AGV.

6. The method of claim 5, wherein the method further comprises: The step 1 comprises: 1.1) laser map construction: placing the AGV on the bottom of the vehicle to be detected, allowing the AGV to explore randomly on the vehicle bottom and construct a laser map, and obtaining a vehicle bottom laser map containing four rectangles representing tires by scanning tire contour point cloud in real time; 1.2) Image map construction: AGV simultaneously collects chassis images in real time through the USB camera with vertical upward view; detects SURF feature points in real-time images, performs FLANN feature matching on SURF feature points of adjacent frame images, calculates homography matrix between adjacent frame images ; according to the homography matrix H, maps and splices the overlapping part of adjacent frame images, and finally splices into a complete chassis image map; 1.3) Map calibration: according to the coordinate matrix of the four tire centers of the left front tire, the right front tire, the left rear tire and the right rear tire in the laser and pixel coordinate system, the coordinate conversion matrix between the laser map and the image map is calculated to complete the map calibration ; 1.4) vehicle bottom positioning correction: performing coordinate system conversion and outlier filtering on the positioning results of the image map and the laser map to correct the vehicle bottom positioning result.

7. The method of claim 5, wherein the method further comprises: The step 2 comprises: 2.1) adaptive histogram equalization: performing adaptive histogram equalization on the original chassis image collected by the camera; 2.2) downsampling: performing downsampling operation on the RGB three channels of the chassis image respectively; this operation can blur the high-frequency details of the parts and highlight the part contour curve; 2.3) Gaussian blur: performing Gaussian blur operation on the chassis image; this operation can blur the high-frequency details of the parts and highlight the part contour curve.

8. The method of claim 5, wherein the method further comprises: The step 5 comprises: 5.1) HSV color space filtering: converting the image from the RGB color space to the HSV color space and filtering based on a preset threshold; 5.2) Component corrosion level evaluation: Perform bitwise AND operations between different component areas and the corrosion area mask to obtain the number of non-zero pixels. Based on the total number of pixels in the part area Calculate the overlap rate ; 5.3) part corrosion sub-region segmentation: performing minimum rectangular bounding box segmentation on the corrosion area with a corrosion coverage exceeding a set threshold, facilitating the review by the corrosion review module.

9. The method of claim 5, wherein the method further comprises: The step 6 comprises: 6.1) strong edge suppression: performing Canny edge detection with different parameters on the part corrosion sub-region and obtaining a corrosion edge map by subtraction; 6.2) local entropy corrosion review: performing average local entropy calculation on the part corrosion detail edge map, and reviewing the corrosion sub-region with an average local entropy greater than a set threshold as having corrosion; The local entropy expression The corrosion threshold value is selected as Wherein: The average value of the local entropy of the image.

10. The method of claim 5, wherein the method further comprises: The step 7 comprises: 7.1) Neural network binary classification review: Through the pre-trained neural network binary classification model, the part corrosion sub-region is reviewed for corrosion / non-corrosion binary classification; The neural network binary classification model is VGG16 convolutional neural network, the sigmoid function is the activation function, and the binary cross-entropy is the loss function; 7.2) Corrosion review result fusion: Perform bit OR operation on the local entropy and the neural network corrosion review result. As long as one of them is judged to exist corrosion, it is finally judged to exist corrosion.

11. The method of claim 5, wherein: The step 8 comprises: 8.1) Corrosion area marking: The part corrosion sub-region, the corrosion evaluation result and the corrosion review result are combined with the corrected vehicle bottom positioning coordinates, and are marked on the image map as the final corrosion evaluation report; 8.2) Corrosion area image saving: The part image judged to have serious corrosion is saved locally on the AGV for reference by the maintenance personnel.

Images