Non-contact tire deformation intelligent identification method based on deep learning

Abstract

The invention discloses a non-contact tire deformation intelligent identification method based on deep learning. The method comprises the following steps: firstly, performing distortion correction on an image acquired by a high-speed optical camera through camera calibration parameters; and then, a target detection and semantic segmentation algorithm is constructed based on a deep learning technology, and tire side images in different environments are trained to obtain an efficient, stable and accurate tire intelligent identification and segmentation model. According to the model, accurate segmentation of a tire to be detected can be realized, geometric fitting is carried out on a mask contour obtained through segmentation through OpenCV, and finally, deformation parameters of the tire are calculated based on a fitted circle. According to the method, pixel-level, stable and high-precision segmentation of the tire deformation area can be realized, the recognition efficiency and accuracy are remarkably improved, and the problems that an existing machine vision recognition method is low in intelligent level and poor in recognition performance in a complex environment are solved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a non-contact tire deformation intelligent recognition method based on deep learning. Background Art

[0002] To ensure the safety of domestic production and transportation, accurately identifying vehicle loads has become a critical issue that needs to be addressed. In recent years, non-contact measurement methods based on machine vision have gained increasing popularity. These methods analyze tire images to extract deformation parameters, thereby inferring vehicle loads, providing a novel approach to load identification. Compared to traditional methods, this technology offers advantages such as non-contact, low cost, and ease of deployment, and has already seen initial application in practice.

[0003] However, current machine vision-based tire deformation measurement technology still faces numerous challenges. For example, high-speed cameras are prone to optical distortion when capturing images, and the image feature extraction process is easily affected by lighting variations and image quality, resulting in unstable measurement accuracy. Furthermore, this technology often relies on complex processing flows and a high level of manual intervention, resulting in limited intelligence and generally applicable only to automated processing in specific scenarios.

[0004] With the rapid development of deep learning technology, it has demonstrated outstanding performance in image processing, with excellent adaptability and generalization capabilities. Leveraging the advantages of deep learning algorithms, it is expected that a more intelligent and stable non-contact tire deformation recognition method will be developed, improving the system's recognition accuracy and robustness in complex environments. Therefore, how to effectively integrate deep learning technology into tire deformation recognition and promote its development towards higher levels of intelligence and universality has become an important topic worthy of in-depth research in this field. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the shortcomings of the above-mentioned existing technologies and provide a non-contact tire deformation intelligent recognition method based on deep learning. This non-contact tire deformation intelligent recognition method based on deep learning can solve the problems of low intelligence level and poor recognition performance in complex environments of the existing method of tire deformation parameter recognition based on machine vision, and realize pixel-level, stable and high-precision segmentation of the deformation area in the tire image, thereby significantly improving the recognition efficiency and accuracy.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A non-contact tire deformation intelligent recognition method based on deep learning, comprising the following steps:

[0008] Step 1: calibrate the parameters of the camera used to capture tire images;

[0009] Step 2: Using the camera calibrated in step 1, capture the side image of the tire to be inspected, and using the calibration parameters in step 1, perform distortion correction on the captured side image of the tire to be inspected to obtain a true side image of the tire;

[0010] Step 3: Input the tire image in step 2 into the pre-trained tire positioning model to obtain the tire position and tire detection frame size;

[0011] Step 4: crop the tire image according to the tire detection frame size obtained in step 3, and then input it into the pre-trained tire segmentation model to obtain the tire mask and wheel mask;

[0012] Step 5: Use OpenCV-based geometric fitting to post-process the tire mask and wheel mask obtained in step 4 to calculate the tire deformation parameters.

[0013] Furthermore, in step 3, YOLOv11 is used as the tire positioning model, and the loss function includes bounding box loss, classification loss, and confidence loss;

[0014] Among them, the bounding box loss adopts CIoU Loss, and its calculation formula is:

[0015]

[0016] Where: w tr 、h tr 、b tr are the width, height, and coordinates of the true bounding box respectively; w, h, b are the width, height, and coordinates of the predicted bounding box respectively; IoU is the intersection over union of the predicted set and the true set; ρ 2 is the square of the Euclidean distance between the center point of the predicted box and the real box; c is the diagonal length of the minimum circumscribed rectangle containing the two boxes; 1 i is an indicator function, which is 1 if the i-th bounding box matches the real target, otherwise it is 0;

[0017] Both classification loss and confidence loss use binary cross entropy loss combined with Focal Loss, and the calculation formula is:

[0018] BCE(p tr,i ,p i )=-[p tr ·log(p)+(1-p tr )·log(1-p)]

[0019] Focal Loss(p,p tr )=-η·(1-p) γ ·BCE(p,p tr )

[0020]

[0021] Where: p tr and p are the probabilities of the true category label and the predicted category label respectively; η is the positive and negative sample balance factor; γ is the weight for controlling difficult and easy samples, which is usually set to 2.

[0022] Furthermore, in step 4, YOLOv11 is used as the tire segmentation model. In addition to the bounding box loss, classification loss, and confidence loss, the loss function also adds a segmentation mask loss, using binary cross entropy loss and Dice Loss. The calculation formula is:

[0023] BCE(m tr,i ,m i )=-[m tr ·log(m)+(1-m tr )·log(1-m)]

[0024]

[0025] Loss4=λ BCE ·L BCE +λ Dice ·L Dice

[0026] Where: m and M are the probability of predicted mask and predicted mask area respectively; m tr and M tr are the probability of the true mask and the true mask area, respectively, λ BCE ,λ Dice is a hyperparameter, and i is the i-th sample.

[0027] Furthermore, in step 4, according to the tire detection frame size obtained in step 3, the aspect ratio of the detection frame is kept unchanged, the tire detection frame size is enlarged (the magnification range is between 1.0 times and 1.5 times), the tire image in step 2 is cropped, and then input into the pre-trained tire segmentation model to obtain the tire mask and wheel mask.

[0028] Furthermore, in step 5, the tire deformation parameters include length type parameters and area type parameters, wherein the length type parameters include tire radius tire_r_real, tire deflection deflection_real, and tire-ground contact line length l_real; the area type parameters include tire area tire_A_real and tire area after compression tire_A_remain.

[0029] Furthermore, the tire deformation parameter calculation method in step 5 includes the following specific steps:

[0030] Step 5-1, performing a Boolean operation on the tire mask and wheel hub mask obtained by segmentation in step 4 to obtain an annular mask area of ​​the tire rubber body;

[0031] Step 5-2, obtaining the fitting circles of the inner and outer contours of the annular mask area of ​​the tire rubber body, specifically includes the following steps:

[0032] Step 5-2A, use the cv2.findContours function to extract the inner and outer contours of the tire rubber body annular mask respectively;

[0033] Step 5-2B, using the least squares algorithm to perform circular fitting on the pixel points on the inner and outer contours of the tire rubber body annular mask region obtained in step 5-2A, and then obtain the pixel coordinates (x) of the center of the inner contour fitting circle of the tire rubber body annular mask region. n ,y n ) pixel , pixel radius rim_r_pixel, and the pixel coordinates of the center of the outer contour fitting circle (x w ,y w ) pixel , pixel radius tire_r_pixel;

[0034] Step 5-3, calculating tire pixel deformation parameters, specifically includes the following steps:

[0035] Step 5-3A, the tire pixel radius is the pixel radius tire_r_pixel of the outer contour fitting circle of the annular mask area of ​​the tire rubber body in step 5-2B;

[0036] In step 5-3B, the wheel hub pixel radius is the pixel radius rim_r_pixel of the contour fitting circle within the annular mask area of ​​the tire rubber body in step 5-2B;

[0037] Step 5-3C, calculate the tire pixel deflection deflection_pixel, the calculation formula is:

[0038] deflection_pixel=y n -y w

[0039] Where y n and y w They are the pixel ordinates of the center of the contour fitting circle in the annular mask area of ​​the tire rubber body and the pixel ordinates of the center of the outer contour fitting circle;

[0040] Step 5-3D: Fit the outer contour of the tire rubber annular mask area obtained in step 5-2B to the circle center pixel y wAdd its pixel radius tire_r_pixel and subtract the tire pixel deflection deflection_pixel obtained in step 5-3C to obtain the specific location of the ground line within the fitted circle of the outer contour of the tire rubber annular mask area. Substitute this into the coordinate formula of the fitted circle of the outer contour of the tire rubber annular mask area to obtain the pixel length l_pixel of the tire-ground contact.

[0041] Step 5-3E, calculate the tire pixel area type parameters, where the tire pixel area tire_A_pixel is calculated using the tire pixel radius tire_r_pixel, and the compressed tire pixel area tire_A_remain_pixel is calculated using the cv2.contourArea function on the tire mask area;

[0042] Step 5-4, introduce the tire deformation proportional factor λ;

[0043] In step 5-5, the tire deformation scale factor λ obtained in step 5-4 is used to correct the tire pixel deformation parameter obtained in step 5-3 to obtain the tire deformation parameter. For length type parameters, the length type pixel parameter is multiplied by the tire deformation scale factor λ during correction. For area type data, the area type pixel parameter is multiplied by the square of the tire deformation scale factor λ during correction.

[0044] Furthermore, in step 5-4, the specific calculation formula of the tire deformation proportional factor λ is:

[0045]

[0046] Where rim_r_real is the real radius of the wheel hub.

[0047] The present invention has the following beneficial effects:

[0048] 1. The present invention corrects the distortion of the monocular camera image and effectively corrects the measurement error.

[0049] 2. The deep learning-based, non-contact, intelligent tire deformation recognition method proposed in this paper integrates object detection and semantic segmentation algorithms within a deep learning framework. Trained on a diverse dataset of tire images constructed under various lighting conditions and complex application environments, the resulting model demonstrates excellent generalization capabilities, enabling efficient and accurate recognition and pixel-level segmentation of diverse tire image types. This method has significantly improved the accuracy and robustness of tire deformation recognition in complex environments in practical machine vision applications.

[0050] 3. The non-contact tire deformation intelligent recognition method based on deep learning proposed in this invention builds a lightweight model for tire deformation parameter recognition based on deep learning training. It can achieve millisecond-level automatic reasoning of tire images in different application environments, greatly improving recognition efficiency and response speed. This method has demonstrated high intelligence and practicality in the field of machine vision, effectively promoting the implementation of tire deformation recognition technology in practical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flowchart of a non-contact tire deformation intelligent recognition method based on deep learning according to an example of the present invention.

[0052] Figure 2 This is a network architecture diagram for the target detection algorithm and semantic segmentation algorithm built based on deep learning.

[0053] Figure 3 Tire detection boxes detected by the tire positioning model.

[0054] Figure 4 Tire mask and wheel mask segmented for the tire segmentation model.

[0055] Figure 5 It is the annular mask of the tire rubber body after Boolean operation.

[0056] Figure 6 is a fitted circle based on the inner and outer contours of the rubber body's annular mask region. DETAILED DESCRIPTION

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and specific preferred embodiments.

[0058] like Figure 1 As shown, a non-contact tire deformation intelligent recognition method based on deep learning includes the following steps:

[0059] Step 1: calibrate the parameters of the camera used to capture tire images;

[0060] Step 2: Using the camera calibrated in step 1, capture the side image of the tire to be inspected, and using the calibration parameters in step 1, perform distortion correction on the captured side image of the tire to be inspected to obtain a true side image of the tire;

[0061] Step 3: Input the tire image in step 2 into the pre-trained tire positioning model to obtain the tire position and tire detection frame size;

[0062] Step 4: crop the tire image according to the tire detection frame size obtained in step 3, and then input it into the pre-trained tire segmentation model to obtain the tire mask and wheel mask;

[0063] Step 5: Use OpenCV-based geometric fitting to post-process the tire mask and wheel mask obtained in step 4 to calculate the tire deformation parameters.

[0064] In an optional embodiment, the camera parameter calibration in step 1 adopts a checkerboard calibration method, and the specific steps are as follows:

[0065] Step 1-1: Print a black and white checkerboard with 9×6 inner corner points. Paste the checkerboard onto a flat board, trying to prevent bending and deformation.

[0066] In steps 1-2, use the camera to be calibrated to capture images of the checkerboard from multiple angles and distances. Collect at least 30 images, and the entire checkerboard should be clearly visible in the images. In this example, the camera to be calibrated is a Sony Alpha ILCE-7M4 SLR camera.

[0067] In steps 1-3, use an OpenCV-based corner detection algorithm to extract the corners of the black and white chessboard in each captured image. In this example, the cv2.findChessboardCorners function is used to extract the corners, and the cv2.cornerSubPix function is used to perform sub-pixel precision processing on the corner points.

[0068] Steps 1-4 use an OpenCV-based camera calibration algorithm to calculate the camera parameter matrix for the extracted black and white checkerboard corner points. In this embodiment, the cv2.calibrateCamera function is used to calculate the camera parameter matrix, and the cv2.FileStorage function is used to save the camera parameter matrix.

[0069] In an optional example, YOLOv11 is used as the tire positioning model in step 3, and the network architecture is as follows: Figure 2 As shown, its loss function includes bounding box loss, classification loss and confidence loss;

[0070] Among them, the bounding box loss adopts CIoU Loss, and its calculation formula is:

[0071]

[0072] Where: w tr 、h tr 、b tr are the width, height, and coordinates of the true bounding box respectively; w, h, b are the width, height, and coordinates of the predicted bounding box respectively; IoU is the intersection over union of the predicted set and the true set; ρ 2 is the square of the Euclidean distance between the center point of the predicted box and the real box; c is the diagonal length of the minimum circumscribed rectangle containing the two boxes; 1i is an indicator function, which is 1 if the i-th bounding box matches the real target, otherwise it is 0;

[0073] Both classification loss and confidence loss use binary cross entropy loss combined with Focal Loss, and the calculation formula is:

[0074] BCE(p tr,i ,p i )=-[p tr ·log(p)+(1-p tr )·log(1-p)]

[0075] Focal Loss(p,p tr )=-η·(1-p) γ ·BCE(p,p tr )

[0076]

[0077] Where: p tr and p are the probabilities of the true category label and the predicted category label respectively; η is the positive and negative sample balance factor; γ is the weight for controlling difficult and easy samples, which is usually set to 2.

[0078] In an optional example, the tire positioning model pre-trained based on YOLOv11 in step 3 is collected and produced by collecting a total of 3494 data sets, which are randomly divided into training set, test set and validation set in a ratio of 7:2:1. The initial learning rate is set to 0.01, and 100 rounds of training are performed. The average accuracy of the tire positioning model obtained on the test set is 0.994.

[0079] In an optional example, YOLOv11 is used as the tire segmentation model in step 4. In addition to the bounding box loss, classification loss, and confidence loss, its loss function also adds a segmentation mask loss, using binary cross entropy loss and DiceLoss, and its calculation formula is:

[0080] BCE(m tr,i ,m i )=-[m tr ·log(m)+(1-m tr )·log(1-m)]

[0081]

[0082] Loss4=λ BCE ·L BCE +λ Dice ·L Dice

[0083] Where: m and M are the probability of predicted mask and predicted mask area respectively; m tr and M tr are the probability of the true mask and the true mask area, respectively.

[0084] In an optional example, the tire segmentation model pre-trained based on YOLOv11 in step 4 is used to collect and produce a total of 3494 data sets, which are randomly divided into training set, test set, and validation set in a ratio of 7:2:1. The initial learning rate is set to 0.01, and 100 rounds of training are performed. The average accuracy of the tire positioning model obtained on the test set is 0.996.

[0085] In an optional embodiment, in step 4, according to the tire detection frame size detected in step 3, as Figure 3 As shown in , keep the aspect ratio of the detection frame unchanged, enlarge the tire detection frame size by 1.2 times, crop the tire image in step 2, and then input the pre-trained tire segmentation model to obtain the tire mask and wheel mask, as shown in Figure 4 shown.

[0086] In an optional example, in step 5, the tire deformation parameters include length type parameters and area type parameters, wherein the length type parameters include tire radius tire_r_real, tire deflection deflection_real, and tire-ground contact line length l_real; the area type parameters include tire area tire_A_real and compressed tire area tire_A_remain.

[0087] In an optional embodiment, the tire deformation parameter calculation method in step 5 includes the following specific steps:

[0088] Step 5-1, perform Boolean operation on the tire mask and wheel mask obtained by segmentation in step 4 to obtain the annular mask area of ​​the tire rubber body, such as Figure 5 As shown;

[0089] Step 5-2, obtaining the fitting circles of the inner and outer contours of the annular mask area of ​​the tire rubber body, specifically includes the following steps:

[0090] Step 5-2A, use the cv2.findContours function to extract the inner and outer contours of the tire rubber body annular mask respectively;

[0091] Step 5-2B, using the least squares algorithm to perform circular fitting on the pixel points on the inner and outer contours of the annular mask area of ​​the tire rubber body obtained in step 5-2A, such as Figure 6 As shown, the pixel coordinates of the center of the inner contour fitting circle of the tire rubber body annular mask area (x n ,yn ) pixel is (2738.97, 2792.27), the pixel radius rim_r_pixel is 735.69, and the pixel coordinates of the center of the outer contour fitting circle (x w ,y w ) pixel is (2727.65, 2783.29), and the pixel radius tire_r_pixel is 1230.30;

[0092] Step 5-3, calculating tire pixel deformation parameters, specifically includes the following steps:

[0093] In step 5-3A, the tire pixel radius is the pixel radius tire_r_pixel of the outer contour fitting circle of the annular mask area of ​​the tire rubber body in step 5-2B, which is 1230.30;

[0094] In step 5-3B, the wheel hub pixel radius is the pixel radius rim_r_pixel of the contour fitting circle in the annular mask area of ​​the tire rubber body in step 5-2B, which is 735.69;

[0095] Step 5-3C, calculate the tire pixel deflection deflection_pixel, the calculation formula is:

[0096] deflection_pixel=y n -y w

[0097] Where y n and y w They are the vertical coordinates of the center pixel of the contour fitting circle within the annular mask area of ​​the tire rubber body and the vertical coordinates of the center pixel of the outer contour fitting circle. The calculated deflection_pixel is 8.98;

[0098] Step 5-3D: Fit the outer contour of the tire rubber annular mask area obtained in step 5-2B to the circle center pixel y w Add its pixel radius tire_r_pixel and subtract the tire pixel deflection deflection_pixel obtained in step 5-3C to obtain the specific location of the ground line within the fitted circle of the outer contour of the tire rubber annular mask area. Substitute this into the coordinate formula of the fitted circle of the outer contour of the tire rubber annular mask area to obtain the tire-ground contact pixel length l_pixel, which is 296.75.

[0099] Step 5-3E, calculate the tire pixel area type parameters, where the tire pixel area tire_A_pixel is calculated using the tire pixel radius tire_r_pixel, and the compressed tire pixel area tire_A_remain_pixel is calculated using the cv2.contourArea function on the tire mask area. The calculated tire_A_pixel is 4755231.45, and the calculated tire_A_remain_pixel is 4753453.73;

[0100] Step 5-4 introduces the tire deformation proportional factor λ, and its specific calculation formula is:

[0101]

[0102] Where rim_r_real is the real radius of the wheel hub, and the calculated proportional factor λ is 0.276;

[0103] In step 5-5, the tire pixel deformation parameters obtained in step 5-3 are corrected using the tire deformation scaling factor λ obtained in step 5-4 to obtain tire deformation parameters. For length type parameters, the length type pixel parameters are multiplied by the tire deformation scaling factor λ during correction. For area type data, the area type pixel parameters are multiplied by the square of the tire deformation scaling factor λ during correction. The tire deformation parameters finally calculated are:

[0104] The tire radius tire_r_real is 339.81 mm, the tire deflection deflection_real is 2.48 mm, the tire-ground contact line length l_real is 81.96 mm, and the area type parameters include the tire area tire_A_real of 362767.99 square millimeters and the compressed tire area tire_A_remain of 362632.37 square millimeters.

[0105] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A non-contact tire deformation intelligent recognition method based on deep learning, characterized by: The following steps are involved: Step 1: calibrate the parameters of the camera used to capture tire images; Step 2: Using the camera calibrated in step 1, capture the side image of the tire to be inspected, and using the calibration parameters in step 1, perform distortion correction on the captured side image of the tire to be inspected to obtain a true side image of the tire; Step 3: Input the tire image in step 2 into the pre-trained tire positioning model to obtain the tire position and tire detection frame size; Step 4: crop the tire image according to the tire detection frame size obtained in step 3, and then input it into the pre-trained tire segmentation model to obtain the tire mask and wheel mask; Step 5: Use OpenCV-based geometric fitting to post-process the tire mask and wheel mask obtained in step 4 to calculate the tire deformation parameters.

2. The non-contact tire deformation intelligent recognition method based on deep learning according to claim 1, characterized in that: In step 3, YOLOv11 is used as the tire positioning model, and the loss function includes bounding box loss, classification loss, and confidence loss; Among them, the bounding box loss adopts CIoU Loss, and its calculation formula is: Where: w tr 、h tr 、b tr are the width, height, and coordinates of the true bounding box respectively; w, h, b are the width, height, and coordinates of the predicted bounding box respectively; IoU is the intersection over union of the predicted set and the true set; ρ 2 is the square of the Euclidean distance between the center point of the predicted box and the real box; c is the diagonal length of the minimum circumscribed rectangle containing the two boxes; 1 i is an indicator function, which is 1 if the i-th bounding box matches the real target, otherwise it is 0; Both classification loss and confidence loss use binary cross entropy loss combined with Focal Loss, and the calculation formula is: BCE(p tr,i ,p i )=-[p tr ·log(p)+(1-p tr )·log(1-p)] Focal Loss(p,p tr )=-η·(1-p) γ ·BCE(p,p tr ) Where: p tr and p are the probabilities of the true category label and the predicted category label respectively; η is the positive and negative sample balance factor; γ is the weight for controlling difficult and easy samples.

3. The non-contact tire deformation intelligent recognition method based on deep learning according to claim 2, characterized in that: In step 4, YOLOv11 is used as the tire segmentation model. In addition to the bounding box loss, classification loss, and confidence loss, the loss function also adds segmentation mask loss, using binary cross entropy loss and Dice Loss. The calculation formula is: BCE(m tr,i ,m i )=-[m tr ·log(m)+(1-m tr )·log(1-m)] Loss4=λ BCE ·L BCE +λ Dice ·L Dice Where: m and M are the probability of predicted mask and predicted mask area respectively; m tr and M tr are the probability of the true mask and the true mask area, respectively, λ BCE ,λ Dice is a hyperparameter, and i is the i-th sample.

4. The non-contact tire deformation intelligent recognition method based on deep learning according to claim 1, characterized in that: In step 4, the tire detection frame size is enlarged according to the tire detection frame size obtained in step 3, while keeping the aspect ratio of the detection frame unchanged. The tire image in step 2 is cropped and then input into the pre-trained tire segmentation model to obtain the tire mask and the wheel mask.

5. The non-contact tire deformation intelligent recognition method based on deep learning according to claim 1, characterized in that: In step 5, the tire deformation parameters include length type parameters and area type parameters, wherein the length type parameters include tire radius tire_r_real, tire deflection deflection_real, and tire-ground contact line length l_real; the area type parameters include tire area tire_A_real and compressed tire area tire_A_remain.

6. The non-contact tire deformation intelligent recognition method based on deep learning according to claim 1, characterized in that: The tire deformation parameter calculation method in step 5 is specifically as follows: Step 5-1, performing a Boolean operation on the tire mask and wheel hub mask obtained by segmentation in step 4 to obtain an annular mask area of ​​the tire rubber body; Step 5-2, obtaining the fitting circles of the inner and outer contours of the annular mask area of ​​the tire rubber body, specifically includes the following steps: Step 5-2A, use the cv2.findContours function to extract the inner and outer contours of the tire rubber body annular mask respectively; Step 5-2B, using the least squares algorithm to perform circular fitting on the pixel points on the inner and outer contours of the tire rubber body annular mask region obtained in step 5-2A, and then obtain the pixel coordinates (x) of the center of the inner contour fitting circle of the tire rubber body annular mask region. n ,y n ) pixel , pixel radius rim_r_pixel, and the pixel coordinates of the center of the outer contour fitting circle (x w ,y w ) pixel , pixel radius tire_r_pixel; Step 5-3, calculating tire pixel deformation parameters, specifically includes the following steps: Step 5-3A, the tire pixel radius is the pixel radius tire_r_pixel of the outer contour fitting circle of the annular mask area of ​​the tire rubber body in step 5-2B; In step 5-3B, the wheel hub pixel radius is the pixel radius rim_r_pixel of the contour fitting circle within the annular mask area of ​​the tire rubber body in step 5-2B; Step 5-3C, calculate the tire pixel deflection deflection_pixel, the calculation formula is: deflection_pixel=y n -y w Where y n and y w They are the pixel ordinates of the center of the contour fitting circle in the annular mask area of ​​the tire rubber body and the pixel ordinates of the center of the outer contour fitting circle; Step 5-3D: Fit the outer contour of the tire rubber annular mask area obtained in step 5-2B to the circle center pixel y w Add its pixel radius tire_r_pixel and subtract the tire pixel deflection deflection_pixel obtained in step 5-3C to obtain the specific location of the ground line within the fitted circle of the outer contour of the tire rubber annular mask area. Substitute this into the coordinate formula of the fitted circle of the outer contour of the tire rubber annular mask area to obtain the pixel length l_pixel of the tire-ground contact. Step 5-3E, calculate the tire pixel area type parameters, where the tire pixel area tire_A_pixel is calculated using the tire pixel radius tire_r_pixel, and the compressed tire pixel area tire_A_remain_pixel is calculated using the cv2.contourArea function on the tire mask area; Step 5-4, introduce the tire deformation proportional factor λ; In step 5-5, the tire deformation scale factor λ obtained in step 5-4 is used to correct the tire pixel deformation parameter obtained in step 5-3 to obtain the tire deformation parameter. For length type parameters, the length type pixel parameter is multiplied by the tire deformation scale factor λ during correction. For area type data, the area type pixel parameter is multiplied by the square of the tire deformation scale factor λ during correction.

7. The non-contact tire deformation intelligent recognition method based on deep learning according to claim 6, characterized in that: In step 5-4, the specific calculation formula of the tire deformation proportional factor λ is: Where rim_r_real is the real radius of the wheel hub.

Images