A wheel alignment method

Abstract

The application relates to a wheel positioning method and a parking robot thereof. The parking robot is suitable for wheel positioning and comprises a robot body, an fisheye camera arranged on the robot body and used for collecting images of the wheel, a laser radar arranged on the robot body and used for collecting point cloud data of the wheel, and a controller arranged on the robot body, which receives and obtains position information of the wheel according to the images and the point cloud data of the fisheye camera and the laser radar, and drives the robot body to move according to the position information. The application provides a wheel positioning method and a parking robot thereof, which can improve the detection accuracy and robustness of the wheel, make the tire grabbing task more efficient and reliable, and make the carrying of the target vehicle more safe.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and artificial intelligence, and in particular to a parking robot and its wheel positioning method. Background Technology

[0002] As people's living standards improve, the number of cars is increasing daily, and parking is becoming a widespread problem. Adopting intelligent parking solutions can better ensure parking efficiency and safety, improve urban space utilization, and solve the "parking difficulty" problem. Achieving this high-quality development transformation requires promoting the application of intelligent parking technology to achieve automated and unmanned efficient operation, including large-scale deployment of AGVs and upgrades to surrounding equipment and station facilities.

[0003] The intelligent parking robot project addresses the needs for safe, efficient, and unmanned car parking by creating an intelligent guided parking system. One key technology for this system is the robot's ability to successfully grasp and move car tires. This requires environmental perception and close-range detection of the wheel's position and posture. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, this invention proposes a parking robot and its wheel positioning method, which can improve the detection accuracy and robustness of wheels, making tire gripping tasks more efficient and reliable, and making the handling of target vehicles safer.

[0005] Specifically, this invention proposes a parking robot suitable for wheel positioning, the parking robot comprising:

[0006] The robot itself;

[0007] A fisheye camera, mounted on the robot body, is used to capture images of the wheels;

[0008] A lidar unit, mounted on the robot body, is used to collect point cloud data of the wheels.

[0009] A controller, mounted on the robot body, receives and obtains the position information of the wheels based on the images and point cloud data from the fisheye camera and lidar. The controller then drives the robot body to move based on the position information.

[0010] According to one embodiment of the present invention, the fisheye camera is mounted on one side of the robot body along its length, and the mounting height of the fisheye camera is 10-20 cm higher than the upper edge of the wheel.

[0011] According to one embodiment of the present invention, the lidar is installed on the same side as the fisheye camera, and the installation height of the lidar is 30-50cm higher than that of the fisheye camera.

[0012] According to one embodiment of the present invention, it further includes a mounting bracket disposed on the robot body, and the lidar is mounted on the mounting bracket.

[0013] According to one embodiment of the present invention, the output video format of the fisheye camera is H.264, and the field of view is 150°.

[0014] According to one embodiment of the present invention, the lidar is a hemispherical multi-line blind spot lidar with a field of view of 90°×360° and a measurement radius of more than 20m.

[0015] The present invention also provides a wheel positioning method, utilizing the aforementioned parking robot, the wheel positioning method comprising the following steps:

[0016] S1. Acquire images and point cloud data of the wheel using the fisheye camera and lidar respectively, and annotate the true bounding box of the wheel;

[0017] S2. Based on the image of the wheel, the wheel target is detected using a two-stage image target detection network; based on the point cloud data of the wheel, the wheel target is detected using a point cloud 3D target detection network PointNet.

[0018] S3. The detection results of the two-stage image target detection network and the point cloud 3D target detection network PointNet are fused to obtain the position information of the wheel;

[0019] S4. Based on the position information of the wheel, determine whether the parking robot is facing the center of the wheel. If yes, proceed to step S5; if no, drive the robot body to move towards the center of the wheel and proceed to step S1.

[0020] S5, End.

[0021] According to one embodiment of the present invention, before step S1, the fisheye camera and the lidar are installed on the robot body, and the internal parameters of the fisheye camera and the external parameters of the lidar are calibrated.

[0022] According to one embodiment of the present invention, the real bounding box is formed by a horizontal line passing through the highest point of the wheel arch, a vertical line passing through the leftmost point of the wheel arch, a vertical line passing through the rightmost point of the wheel arch, and a horizontal line passing through the ground contact point of the wheel.

[0023] According to an embodiment of the present invention, in step S2, detecting the wheel target through the two-stage image target detection network includes the following steps:

[0024] S201. Input the entire image into the pre-trained deep convolutional network and extract the corresponding feature maps;

[0025] S202. Input the feature map into the region generation network. Each position on the feature map corresponds to 9 anchor points on the original image. The region generation network classifies each anchor point to determine whether it is a target or background. The network also performs preliminary regression of the bounding box position for the selected targets to generate target candidate boxes.

[0026] S203. Project the target candidate boxes generated by the region generation network onto the feature map to obtain the corresponding feature matrix;

[0027] S204. Scale each feature matrix to 7x7 using a region-of-interest pooling layer. 7. Size;

[0028] S205, 7 The feature matrix of size 7 is flattened, input into a series of fully connected layers, and the candidate region classification results are obtained, and the target bounding box of the wheel is located.

[0029] According to an embodiment of the present invention, in step S201, the training process of the deep convolutional network adopts the joint training method of the region generation network multi-task loss function and the fully connected layer multi-task loss function, and at the same time trains the weight parameters of the entire two-stage image object detection network architecture.

[0030] According to an embodiment of the present invention, in step S2, detecting the wheel target using the PointNet point cloud 3D target detection network includes the following steps:

[0031] S211, For each N The point cloud input is 3. The point cloud 3D target detection network first learns the rotation of the point cloud through a T-Net, aligns it in space, then maps it to a 64-dimensional space through a multilayer perceptron, then aligns it through a T-Net, and finally maps it to a 1024-dimensional space through a multilayer perceptron.

[0032] S212, Regarding N obtained in step S211 A 1024-dimensional vector representation is used, and a max-pooling operation is introduced to retain only the largest number across all channels of the 1024-dimensional vector, resulting in 1. Global features of a 1024-dimensional vector;

[0033] S213. Concatenate the global features to the point-by-point features of the 64-dimensional point cloud in step 211 to obtain N. The 1088-dimensional vector representation is used, and finally, a multilayer perceptron is used to output the classification probability of each point, thus completing the point cloud segmentation and detection of the wheel target.

[0034] According to an embodiment of the present invention, in step S3, based on the calibrated internal parameters of the fisheye camera and the external parameters of the lidar, the true bounding box of the wheel obtained in step S1 is mapped from the image coordinate system to the lidar coordinate system. Then, the point cloud segmentation detection result obtained in step S2 is used to verify the image detection result, and the detection results of the image and the lidar point cloud are further fused to finally output the position information of the wheel.

[0035] According to one embodiment of the present invention, the position information of the wheel includes the three-dimensional coordinates of the wheel's center, the wheel size, and the wheel deflection angle.

[0036] This invention provides a parking robot and its wheel positioning method, which is based on the fusion detection of fisheye camera and lidar. This method can improve the detection accuracy and robustness of wheels, making tire grasping tasks more efficient and reliable, and making the handling of target vehicles safer. This research has important practical application value.

[0037] It should be understood that the above general description and the following detailed description of the present invention are exemplary and illustrative, and are intended to provide further explanation of the present invention. Attached Figure Description

[0038] The accompanying drawings are included to provide a further understanding of the invention; they are incorporated into and constitute a part of this application. The drawings illustrate embodiments of the invention and, together with this specification, serve to explain the principles of the invention. In the drawings:

[0039] Figure 1 A schematic diagram of the structure of a parking robot according to an embodiment of the present invention is shown.

[0040] Figure 2 yes Figure 1 The right view.

[0041] Figure 3 A diagram illustrating the usage state of a parking robot according to an embodiment of the present invention is shown.

[0042] Figure 4 A flowchart of a wheel alignment method according to an embodiment of the present invention is shown.

[0043] Figure 5 A flowchart illustrating a two-stage image target detection network according to an embodiment of the present invention for detecting wheel targets is shown.

[0044] Figure 6 The flowchart illustrates a point cloud 3D target detection network according to an embodiment of the present invention for detecting wheel targets. Detailed Implementation

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0048] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0049] In the description of this application, it should be understood that the orientation or positional relationship indicated by directional terms such as "front, back, up, down, left, right", "horizontal, vertical, horizontal" and "top, bottom" is usually based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing this application and simplifying the description. Unless otherwise stated, these directional terms 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 on the scope of protection of this application; the directional terms "inner" and "outer" refer to the inner and outer contours relative to the outline of each component itself.

[0050] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "on top of" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways, rotated 90 degrees, or in other orientations, and the spatial relative descriptions used herein will be interpreted accordingly.

[0051] Furthermore, it should be noted that the use of terms such as "first" and "second" to define components is merely for the purpose of distinguishing the corresponding components. Unless otherwise stated, these terms have no special meaning and therefore should not be construed as limiting the scope of protection of this application. In addition, although the terminology used in this application is selected from commonly known and used terms, some terms mentioned in this application's specification may have been chosen by the applicant according to his or her judgment, and their detailed meanings are explained in the relevant sections of this description. Moreover, this application should be understood not only through the actual terms used, but also through the meaning implied by each term.

[0052] Figure 1 A schematic diagram of the structure of a parking robot according to an embodiment of the present invention is shown. Figure 2 yes Figure 1 The right view. Figure 3 The figure shows a diagram illustrating the usage state of a parking robot according to an embodiment of the present invention. As shown, a parking robot 100 suitable for wheel positioning mainly includes a robot body 101, a fisheye camera 102, a lidar 103, and a controller 104.

[0053] The fisheye camera 102, lidar 103, and controller 104 are mounted on the robot body 101.

[0054] The fisheye camera 102 is used to capture images of the wheels.

[0055] The lidar 103 is used to collect point cloud data of the wheels.

[0056] The controller 104 receives and obtains the wheel's position information based on the images and point cloud data from the fisheye camera 102 and the lidar 103. Based on this position information, the controller 104 drives the robot body 101 to move, positioning the robot body 101 directly opposite the center of the wheel, preparing it to grasp the wheel and lift the vehicle 200. In other words, the controller 104 uses the fisheye camera 102 and the lidar 103 in conjunction to perform real-time detection and positioning of the wheel under test and the center position of its outer surface, thereby driving the parking robot 100 to a specific working position.

[0057] Better, refer to Figure 2 and Figure 3 A fisheye camera 102 is mounted on one side of the robot body 101 along its length. The mounting height of the fisheye camera 102 is 10-20 cm higher than the upper edge of the wheel. More preferably, refer to... Figure 3 The fisheye camera 102 is tilted downwards at a certain angle to ensure that the area below the robot body 101 is the boundary of its field of view, allowing a complete observation of the ground surrounding the side of the robot body 101 closest to the wheels of the vehicle 200. As is easily understood, the fisheye camera 102 transmits the captured image information to the controller 104 via the ROS2 system. The image from the fisheye camera 102 undergoes wide-angle distortion correction, resulting in less distortion in the final image.

[0058] Preferably, the lidar 103 is mounted on the same side as the fisheye camera 102. The lidar 103 is mounted at a height 30-50 cm higher than the fisheye camera 102. The lidar 103 can be understood as being used to assist in detecting whether the robot body 101 and the wheel being measured are aligned, and to determine the distance between the robot body and the robot body 101.

[0059] Preferably, the parking robot 100 also includes a mounting bracket 105 disposed on the robot body 101, and the lidar 103 is mounted on the mounting bracket 105 so that the lidar 103 has a higher mounting position than the fisheye camera 102.

[0060] Preferably, the fisheye camera 102 outputs video in H.264 format with a field of view of 150°.

[0061] Preferably, the lidar 103 is a hemispherical multi-line blind-spot lidar 103 with a field of view of 90°×360° and a measurement radius exceeding 20m. (Reference) Figure 3 A portion of the lidar 103 is exposed on the side near the wheel to ensure a 180° measurement angle.

[0062] The present invention also provides a wheel positioning method using the aforementioned parking robot 100. Figure 4 A flowchart of a wheel alignment method according to an embodiment of the present invention is shown. As shown, the wheel alignment method includes the following steps:

[0063] S1. Collect images and point cloud data of the wheel using fisheye camera 102 and lidar 103 respectively, and mark the true bounding box of the wheel;

[0064] S2. Based on the image of the wheel, the wheel target is detected by a two-stage image target detection network; based on the point cloud data of the wheel, the wheel target is detected by a point cloud 3D target detection network (PointNet).

[0065] S3. The detection results of the two-stage image target detection network and the point cloud 3D target detection network are fused to obtain the position information of the wheel;

[0066] S4. Based on the position information of the wheel, determine whether the parking robot 100 is facing the center of the wheel. If yes, proceed to step S5; otherwise, drive the robot body 101 to move towards the center of the wheel and proceed to step S1.

[0067] S5, End.

[0068] The present invention provides a wheel positioning method that utilizes a fisheye camera 102 and a lidar 103 to dynamically detect the wheel position. A controller 104 can control a parking robot 100 to continuously detect and iterate as it approaches the detected wheel, forming a closed-loop control. Close-range dynamic detection begins approximately 70cm from the target wheel, causing the parking robot 100 to approach a position directly opposite the center of the outer circumference of the wheel being measured, ultimately reaching a precise positioning point where the parking robot 100 is directly facing the center of the wheel.

[0069] Preferably, before step S1, refer to Figure 1 and Figure 2 A fisheye camera 102 and a lidar 103 are mounted on the robot body 101. The fisheye camera 102 can be mounted at the center of one side of the robot body 101 along its length, where the gripper arm can extend, and tilted downwards at a certain angle, such as 45°, to ensure that the area below the robot body 101 is the boundary of the field of view, allowing for a complete observation of the wheel under test and the surrounding ground. The wide field of view of the fisheye camera 102 enables it to detect wheel targets at close range with a small blind zone, resulting in high detection accuracy and robustness. Through dynamic iteration during close-range detection, it can achieve precise alignment with the center of the wheel. The lidar 103 is mounted on a bracket on the top surface of the robot body 101, near the wheel under test, and is a multi-line high-precision lidar 103, primarily used to detect the three-dimensional coordinates of the center of the wheel under test. Furthermore, before step S1, the internal parameters of the fisheye camera 102 and the external parameters of the lidar 103 are calibrated.

[0070] Preferably, the true bounding box is formed by a horizontal line passing through the highest point of the wheel arch, a vertical line passing through the leftmost point of the wheel arch, a vertical line passing through the rightmost point of the wheel arch, and a horizontal line passing through the point where the wheel touches the ground.

[0071] Figure 5 A flowchart illustrating a two-stage image target detection network for detecting wheel targets according to an embodiment of the present invention is shown. As shown, preferably, in step S2, detecting the wheel target using the two-stage image target detection network includes the following steps:

[0072] S201. Input the entire image into the pre-trained deep convolutional network and extract the corresponding feature maps;

[0073] S202. Input the feature map into the Region Generation Network (PRN). Each position on the feature map corresponds to 9 anchor points on the original image. The Region Generation Network classifies each anchor point to determine whether it is a target or background, and initially regresses the bounding box position of the selected targets to generate target candidate boxes.

[0074] S203. Project the target candidate boxes generated by the region generation network onto the feature map to obtain the corresponding feature matrix;

[0075] S204. Scale each feature matrix uniformly to 7 using a Region of Interest (ROI) pooling layer. 7. Size;

[0076] S205, 7 The feature matrix of size 7 is flattened, input into a series of fully connected layers, and the candidate region classification results are obtained. The target bounding box of the wheel is then located.

[0077] Preferably, in step S201, the training process of the deep convolutional network adopts a joint training method using the multi-task loss function of the region generation network and the multi-task loss function of the fully connected layer, and simultaneously trains the weight parameters of the entire two-stage image object detection network architecture. The wheel localization method of the present invention utilizes a two-stage image object detection network to quickly and accurately select the wheel arch of the tested wheel, and trains it using a model with the known center position of the outer circle of the wheel to learn how to find and locate the center of the outer circular surface of the tested wheel.

[0078] Figure 6 A flowchart illustrating the PointNet point cloud 3D target detection network's detection of wheel targets according to an embodiment of the present invention is shown. As shown, preferably, in step S2, the PointNet point cloud 3D target detection network detects the wheel targets, primarily by classifying the input point cloud data point by point, thereby segmenting the portion of the point cloud identified as wheel targets. This detection includes the following steps:

[0079] S211, For each N The 3D point cloud input is used to learn the rotation of the point cloud through a T-Net and align it in space. Then, it is mapped to a 64-dimensional space through a multilayer perceptron (MLP), aligned again through a T-Net, and finally mapped to a 1024-dimensional space through a multilayer perceptron.

[0080] S212, For N obtained from step S211 A 1024-dimensional vector representation is used, and a max-pooling operation is introduced to retain only the largest number across all channels of the 1024-dimensional vector, resulting in 1. Global features of a 1024-dimensional vector;

[0081] S213. Concatenate the global features to the 64-dimensional point-by-point features of the point cloud from step 211 to obtain N. The 1088-dimensional vector representation is finally passed through a multilayer perceptron to output the classification probability of each point, that is, to determine whether it is the target wheel or the background, thus completing the point cloud segmentation and detection of the wheel target.

[0082] Preferably, in step S3, based on the calibrated internal parameters of the fisheye camera and the external parameters of the LiDAR, the true bounding box of the wheel obtained in step S1 is mapped from the image coordinate system to the LiDAR coordinate system. The point cloud segmentation detection results obtained in step S2 are then used to verify the image detection results, and the detection results of the image and LiDAR point cloud are further fused to finally output the wheel's position information. Based on this wheel position information, the robot's pose is adjusted, facing the wheel plane, aligning with the wheel's center, and moving to the next detection-movement iteration. This process is repeated until precise alignment is achieved.

[0083] Preferably, the wheel's position information includes the wheel's center three-dimensional coordinates, wheel size, and wheel deflection angle.

[0084] It will be apparent to those skilled in the art that various modifications and variations can be made to the exemplary embodiments described above without departing from the spirit and scope of the invention. Therefore, it is intended that this invention cover modifications and variations falling within the scope of the appended claims and their equivalents.

Claims

1. A wheel positioning method, applicable to a parking robot, the parking robot comprising: The robot itself; A fisheye camera, mounted on the robot body, is used to capture images of the wheels; A lidar unit, mounted on the robot body, is used to collect point cloud data of the wheels. A controller, mounted on the robot body, receives and obtains the position information of the wheels based on the images and point cloud data from the fisheye camera and lidar, and drives the robot body to move according to the position information. The wheel alignment method includes the following steps: S1. The image and point cloud data of the wheel are acquired by the fisheye camera and the lidar respectively, and the true boundary box of the wheel is marked. The true boundary box is formed by a horizontal line passing through the highest point of the wheel arch, a vertical line passing through the leftmost point of the wheel arch, a vertical line passing through the rightmost point of the wheel arch, and a horizontal line passing through the ground contact point of the wheel. S2. Based on the image of the wheel, the wheel target is detected using a two-stage image target detection network; based on the point cloud data of the wheel, the wheel target is detected using a point cloud 3D target detection network; wherein, the detection of the wheel target using the two-stage image target detection network includes the following steps: S201. Input the entire image into the pre-trained deep convolutional network and extract the corresponding feature maps; S202. Input the feature map into the region generation network. Each position on the feature map corresponds to 9 anchor points on the original image. The region generation network classifies each anchor point to determine whether it is a target or background. The network also performs preliminary regression of the bounding box position for the selected targets to generate target candidate boxes. S203. Project the target candidate boxes generated by the region generation network onto the feature map to obtain the corresponding feature matrix; S204. Scale each feature matrix to 7x7 using a region-of-interest pooling layer.

7. Size; S205, 7 The feature matrix of size 7 is flattened, input into a series of fully connected layers, and the candidate region classification results are obtained, and the target bounding box of the wheel is located. S3. The detection results of the two-stage image target detection network and the point cloud 3D target detection network are fused to obtain the position information of the wheel; S4. Based on the position information of the wheel, determine whether the parking robot is facing the center of the wheel. If yes, proceed to step S5; if no, drive the robot body to move towards the center of the wheel and proceed to step S1. S5, End.

2. The wheel positioning method as described in claim 1, characterized in that, The fisheye camera is mounted on one side of the robot body along its length, and the mounting height of the fisheye camera is 10-20 cm higher than the upper edge of the wheel.

3. The wheel alignment method as described in claim 2, characterized in that, The lidar is installed on the same side as the fisheye camera, and the installation height of the lidar is 30-50cm higher than that of the fisheye camera.

4. The wheel positioning method as described in claim 1, characterized in that, The parking robot also includes a mounting bracket disposed on the robot body, and the lidar is mounted on the mounting bracket.

5. The wheel positioning method as described in claim 1, characterized in that, The fisheye camera outputs video in H.264 format with a field of view of 150°.

6. The wheel positioning method as described in claim 1, characterized in that, The lidar is a hemispherical multi-line blind spot lidar with a field of view of 90°×360° and a measurement radius of over 20m.

7. The wheel positioning method as described in claim 1, characterized in that, Before step S1, the fisheye camera and lidar are installed on the robot body, and the internal parameters of the fisheye camera and the external parameters of the lidar are calibrated.

8. The wheel alignment method as described in claim 7, characterized in that, In step S201, the training process of the deep convolutional network adopts a joint training method of the region generation network multi-task loss function and the fully connected layer multi-task loss function, and at the same time trains the weight parameters of the entire two-stage image object detection network architecture.

9. The wheel positioning method as described in claim 1, characterized in that, In step S2, the detection of the wheel target using the point cloud 3D target detection network includes the following steps: S211, For each N The point cloud input is 3. The point cloud 3D target detection network first learns the rotation of the point cloud through a T-Net, aligns it in space, then maps it to a 64-dimensional space through a multilayer perceptron, then aligns it through a T-Net, and finally maps it to a 1024-dimensional space through a multilayer perceptron. S212, Regarding N obtained in step S211 A 1024-dimensional vector representation is used, and a max-pooling operation is introduced to retain only the largest number across all channels of the 1024-dimensional vector, resulting in 1. Global features of a 1024-dimensional vector; S213. Concatenate the global features to the point-by-point features of the 64-dimensional point cloud in step 211 to obtain N. The 1088-dimensional vector representation is used, and finally, a multilayer perceptron is used to output the classification probability of each point, thus completing the point cloud segmentation and detection of the wheel target.

10. The wheel positioning method as described in claim 9, characterized in that, In step S3, based on the calibrated internal parameters of the fisheye camera and the external parameters of the lidar, the true bounding box of the wheel obtained in step S1 is mapped from the image coordinate system to the lidar coordinate system. Then, the point cloud segmentation detection results obtained in step S2 are used to verify the image detection results, and the detection results of the image and the lidar point cloud are further fused to finally output the position information of the wheel.

11. The wheel positioning method as described in claim 10, characterized in that, The position information of the wheel includes the three-dimensional coordinates of the wheel's center, the wheel size, and the wheel's deflection angle.

Images