A wall column detection method and a wall column detection system for parking

Abstract

The application provides a wall column detection method and wall column detection system for parking, and relates to the technical field of vehicle parking. The application obtains an aerial view around a vehicle when the vehicle is parking, and then detects the aerial view by using a deep learning detection network to obtain the type of a target wall column in the aerial view and initial grounding data of the target wall column. After the initial grounding data of the target wall column in the aerial view is mapped to a three-dimensional coordinate system, the invisible part of the target wall column in contact with the ground is filled according to the type of the target wall column and the initial grounding data of the target wall column to obtain all grounding data of the target wall column. Finally, parking is performed according to the all grounding data of the target wall column. The application can fill the invisible area in the wall column, output complete grounding information of the wall column, accurately determine the position of the wall column in the three-dimensional coordinate system, provide convenience for parking, and avoid the situation that the wall column is collided in the parking process due to the lack of grounding information of the wall column.

Description

Technical Field

[0001] This invention relates to the field of vehicle parking technology, and in particular to a method and system for detecting wall pillars used in parking. Background Technology

[0002] With the rapid development of autonomous driving technology, various advanced technologies in parking scenarios are gradually being applied to various vehicle models. The detection of obstacles such as walls and pillars during the parking process is becoming increasingly important. Accurate wall and pillar detection results are crucial for obstacle avoidance during automatic parking or driver parking. Furthermore, wall and pillar detection results can be rendered in the vehicle's visualization system to enhance the driver's perception of the surrounding environment and help the driver park or automatically park.

[0003] Currently, there are two main approaches to wall / pillar detection in parking scenarios. One approach involves using a surround-view camera mounted on the vehicle to acquire four fisheye images, labeling the wall / pillar bounding boxes in these four images, and inputting them into a convolutional neural network. This network then combines common anchor-based / anchor-free 2D object detection methods to detect the wall / pillars in each of the four images. However, this method does not consider the computational and resource consumption of detecting four images. If only a single viewpoint image is detected, the system's environmental perception range is significantly limited. Furthermore, traditional 2D bounding box wall / pillar detection methods cannot accurately convert and obtain the wall / pillar positions in 3D space. The other approach uses ultrasonic radar mounted around the vehicle. By using ultrasonic ranging combined with clustering and filtering algorithms, the system can determine the obstacles around the vehicle. This method can effectively compensate for some near-range obstacles that cameras cannot detect, reducing the risk of collisions during parking. However, in real-world parking scenarios, wall / pillars are often covered with advertisements or anti-collision materials. Ultrasonic waves are easily affected by the insufficient reflectivity of the wall / pillar surface, making it difficult to accurately determine the wall / pillar's position, significantly increasing safety hazards. Summary of the Invention

[0004] One objective of the first aspect of this invention is to provide a method for detecting wall pillars for parking, thereby solving the technical problem that the prior art cannot accurately determine the position of wall pillars.

[0005] Another objective of the first aspect of this invention is to reduce the computing power and resource consumption of the system.

[0006] The second aspect of this invention aims to provide a wall column detection system for parking.

[0007] According to the first aspect of the present invention, the present invention also provides a method for detecting wall pillars for parking, comprising the following steps:

[0008] Obtain a bird's-eye view of the area around the vehicle while it is parked;

[0009] The bird's-eye view is detected using a deep learning detection network to obtain the type of the target wall column in the bird's-eye view and the initial grounding data of the target wall column, wherein the initial grounding data represents the data of the visible part of the target wall column in contact with the ground;

[0010] After mapping the initial grounding data of the target wall pillar in the bird's-eye view to the three-dimensional coordinate system in the vehicle, the invisible part of the target wall pillar in contact with the ground is filled in according to the type of the target wall pillar and the initial grounding data of the target wall pillar, so as to obtain the complete grounding data of the target wall pillar.

[0011] Parking is performed based on all grounding data of the target wall column.

[0012] Optionally, the step of using a deep learning detection network to detect the bird's-eye view to obtain the type of the target wall column and the initial grounding data of the target wall column in the bird's-eye view specifically includes the following steps:

[0013] A deep learning detection network is used to detect the bird's-eye view to identify the location of the target wall column and ground information;

[0014] The initial grounding data of the target wall column is determined based on its location and the ground information.

[0015] The type of the target wall column is determined based on the initial grounding data of the target wall column.

[0016] Optionally, the step of determining the type of the target wall column based on the initial grounding data of the target wall column specifically includes the following steps:

[0017] The shape of the visible portion of the target wall column in contact with the ground is determined based on the initial grounding data of the target wall column;

[0018] The type of the target wall column is determined based on the shape of the visible portion of the target wall column in contact with the ground.

[0019] Optionally, the step of determining the type of the target wall column based on the shape of the visible portion of the target wall column in contact with the ground specifically includes the following steps:

[0020] If the visible portion of the target wall column in contact with the ground is arc-shaped, then the type of the target wall column is determined to be cylindrical.

[0021] If the visible portion of the target wall column in contact with the ground is composed of two connected straight lines, then the type of the target wall column is determined to be a quadrangular prism.

[0022] If the visible portion of the target wall in contact with the ground is composed of at least three sequentially connected straight lines, then the target wall column is determined to be a polygonal prism, and the number of sides of the base of the polygonal prism is greater than four.

[0023] Optionally, if the visible portion of the target wall column in contact with the ground consists of two connected straight lines, then determining the type of the target wall column as a quadrangular prism specifically includes the following steps:

[0024] If the included angle between two connecting straight lines is equal to 90°, then the target wall column is determined to be of the shape of a square prism.

[0025] If the angle between two straight lines is less than or greater than 90°, then the target wall column is determined to be a non-square prism, specifically a quadrangular prism.

[0026] Optionally, the step of filling in the invisible portion of the target wall column in contact with the ground based on the type of the target wall column and the initial grounding data of the target wall column to obtain the complete grounding data of the target wall column specifically includes the following steps:

[0027] If the target wall column is cylindrical, then the coordinates of the three grounding points where the target wall column contacts the ground are selected from the initial grounding data of the target wall column;

[0028] The coordinates of the center of the bottom surface of the target wall column in contact with the ground and the radius of the bottom surface are calculated based on the coordinates of the three grounding points.

[0029] The coordinates of all grounding points of the target wall column in contact with the ground are determined based on the coordinates of the center of the bottom surface and the radius of the bottom surface, thereby obtaining all grounding data of the target wall column.

[0030] Optionally, the step of filling in the invisible portion of the target wall column in contact with the ground based on the type of the target wall column and the initial grounding data of the target wall column to obtain the complete grounding data of the target wall column specifically includes the following steps:

[0031] If the target wall column is a quadrangular prism, then the lengths and coordinates of the two straight lines that connect the target wall column to the ground are obtained from the initial grounding data of the target wall column.

[0032] Based on the lengths and coordinates of the two straight lines that connect the target wall column to the ground, the lengths and coordinates of the other two straight lines that connect the target wall column to the ground are determined, thereby obtaining all the grounding data of the target wall column.

[0033] Optionally, the step of filling in the invisible portion of the target wall column in contact with the ground based on the type of the target wall column and the initial grounding data of the target wall column to obtain the complete grounding data of the target wall column specifically includes the following steps:

[0034] If the target wall column is a polygonal prism, then obtain the lengths of at least three straight lines that the target wall column contacts the ground, the angles between two adjacent straight lines, and the coordinates of at least three straight lines from the initial grounding data of the target wall column.

[0035] The length and coordinates of the remaining straight lines in contact with the ground are determined based on the lengths of at least three straight lines in contact with the target wall column, the angle between two adjacent straight lines, and the coordinates of the at least three straight lines, thereby obtaining all the grounding data of the target wall column.

[0036] Optionally, the step of using a deep learning detection network to detect the bird's-eye view to identify the location of the target wall column and ground information specifically includes the following steps:

[0037] Select a feature map from the bird's-eye view;

[0038] Feature layers of different sizes are obtained from the feature map using upsampling and downsampling methods, and then a feature pyramid is formed.

[0039] In the feature pyramid, feature layers of different sizes output semantic segmentation results and instance segmentation results respectively;

[0040] The semantic segmentation results and instance segmentation results of all the feature layers are superimposed to obtain the location of the target wall column and the ground information.

[0041] According to a second aspect of the present invention, the present invention also provides a wall column detection system for parking, comprising:

[0042] The control module includes a memory and a processor. The memory stores a calculation program, which, when executed by the processor, is used to implement the above-described calculation method.

[0043] This invention first acquires a bird's-eye view of the vehicle's surroundings while parking. Then, a deep learning detection network is used to detect the bird's-eye view to obtain the type of target wall pillars and their initial grounding data. Next, the initial grounding data of the target wall pillars in the bird's-eye view is mapped to a three-dimensional coordinate system. Based on the type of target wall pillar and its initial grounding data, the invisible parts of the target wall pillars in contact with the ground are filled in to obtain complete grounding data for the target wall pillars. Finally, parking is performed based on all the grounding data of the target wall pillars. This technical solution can fill in the invisible areas of the wall pillars, outputting complete grounding information, thereby accurately determining the position of the wall pillars in the three-dimensional coordinate system, facilitating parking and preventing collisions with wall pillars during parking due to missing grounding information.

[0044] Furthermore, the present invention performs wall and column detection in a bird's-eye view and shares the main structure with conventional parking space detection in a bird's-eye view. Compared with technical solutions that require the creation of a separate 3D model, this invention can significantly save resources and energy consumption and reduce the computing power of the system.

[0045] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0046] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0047] Figure 1 This is a schematic flowchart of a wall column detection method for parking according to an embodiment of the present invention;

[0048] Figure 2 This is a schematic flowchart of a wall column detection method for parking according to another embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram showing that the visible portion of the target wall column in contact with the ground is arc-shaped, according to one embodiment of the present invention.

[0050] Figure 4 Yes Figure 3 A schematic diagram showing the bottom surface of the target wall column after it has been filled in contact with the ground;

[0051] Figure 5 This is a schematic diagram showing that, according to one embodiment of the present invention, the visible portion of the target wall column in contact with the ground is shaped as two connected straight lines;

[0052] Figure 6Yes Figure 5 A schematic diagram showing the bottom surface of the target wall column after it has been filled in contact with the ground.

[0053] Figure 7 This is a schematic connection block diagram of a wall column detection system for parking according to an embodiment of the present invention. Detailed Implementation

[0054] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0055] Figure 1 This is a schematic flowchart of a wall column detection method for parking according to an embodiment of the present invention. Figure 1 As shown, in this embodiment, the wall column detection method for parking includes the following steps:

[0056] Step S100: Obtain a bird's-eye view of the area around the vehicle while it is parked;

[0057] Step S200: Use a deep learning detection network to detect the bird's-eye view to obtain the type of the target wall column and the initial grounding data of the target wall column in the bird's-eye view. The initial grounding data represents the data of the visible part of the target wall column in contact with the ground.

[0058] Step S300: After mapping the initial grounding data of the target wall pillar in the bird's-eye view to the three-dimensional coordinate system in the vehicle, fill in the invisible part of the target wall pillar in contact with the ground according to the type of the target wall pillar and the initial grounding data of the target wall pillar to obtain all the grounding data of the target wall pillar.

[0059] Step S400: Park the vehicle based on all grounding data of the target wall column.

[0060] This embodiment can fill in the invisible areas of the wall column and output the complete grounding information of the wall column, thereby accurately determining the position of the wall column in the three-dimensional coordinate system, which provides convenience for parking and avoids collisions with the wall column during parking due to the lack of grounding information.

[0061] This embodiment performs wall and column detection on a bird's-eye view, sharing the same backbone as conventional parking space detection on a bird's-eye view. Compared to technical solutions that require building a separate 3D model, this significantly saves resources and energy consumption and reduces system computational burden. Assuming that building a bird's-eye view requires 1 resource, building a new 3D model requires 4 resources, plus the original bird's-eye view, totaling 5 resources. Therefore, this embodiment performs wall and column detection on the existing bird's-eye view, saving 4 resources.

[0062] In step S100, the four-way surround view fisheye images of the vehicle are converted and stitched together to form a bird's-eye view based on the vehicle's location center.

[0063] Figure 2 This is a schematic flowchart of a wall column detection method for parking according to another embodiment of the present invention. Figure 2 As shown, in this embodiment, step S200 specifically includes the following steps:

[0064] Step S210: Use a deep learning detection network to detect the bird's-eye view to identify the location of the target wall column and ground information;

[0065] Step S220: Determine the initial grounding data of the target wall column based on its location and ground information;

[0066] Step S230: Determine the type of the target wall column based on the initial grounding data of the target wall column.

[0067] In step S230, the main consideration is that if the target wall column is cylindrical overall, but for design purposes, the section in contact with the ground might be square. Imagine only the section connected to the ground is square, with a large portion above it being cylindrical. Directly determining the shape of the target wall's contact with the ground as circular based solely on the general shape of the target wall in a bird's-eye view might be inaccurate and affect the accuracy of the wall's grounding information. Therefore, this embodiment directly determines the target wall's type based on its initial grounding data, thereby improving the accuracy of the wall's grounding information determination.

[0068] In this embodiment, step S230 specifically includes the following steps:

[0069] Step S231: Determine the shape of the visible part of the target wall column in contact with the ground based on the initial grounding data of the target wall column;

[0070] Step S232: Determine the type of target wall column based on the shape of the visible part of the target wall column in contact with the ground.

[0071] In this embodiment, step S232 specifically includes the following steps:

[0072] Step S2321: If the visible part of the target wall column in contact with the ground is arc-shaped, then the type of the target wall column is determined to be cylindrical.

[0073] Step S2322: If the visible part of the target wall column in contact with the ground is two connected straight lines, then the type of the target wall column is determined to be a quadrangular prism.

[0074] In step S2323, if the visible portion of the target wall in contact with the ground consists of at least three consecutively connected straight lines, then the target wall column is determined to be a polygonal prism, and the number of sides of the base of the polygonal prism is greater than four. It should be noted that steps S2321, S2322, and S2323 are not sequential.

[0075] Step S2322 specifically includes the following steps:

[0076] Step 1: If the angle between the two connecting straight lines is 90°, then the target wall column is determined to be a square prism. Here, the bottom surface of the square prism target wall column that contacts the ground is rectangular.

[0077] Step two: If the angle between the two straight lines is less than or greater than 90°, then the target wall column is determined to be a non-square prism, specifically a quadrangular prism. It should be noted that there is no specific order between steps one and two. The base of a non-square prism, quadrangular prism target wall column that contacts the ground is a non-rectangular parallelogram.

[0078] In step two, when the angle between the two straight lines is less than 90°, the target wall column could be a parallelogram or a triangle, but it is considered to be a quadrangular prism. In other words, this embodiment uses a larger-than-smaller approach to determine the type of the target wall column. If the base of the target wall column in contact with the ground is triangular, it is actually a triangular prism. Considering it a quadrangular prism avoids collisions between the vehicle and the wall column during parking, maximizing the accuracy of the wall column grounding information determination while ensuring safety.

[0079] Figure 3 This is a schematic diagram illustrating that, according to one embodiment of the present invention, the visible portion of the target wall column in contact with the ground is arc-shaped. Figure 4 Yes Figure 3 A schematic diagram showing the bottom surface after filling where the target wall column contacts the ground, as shown below. Figure 3 and Figure 4 As shown, the solid lines represent the visible portion of the target wall, and the dashed lines represent the invisible portion of the target wall where it meets the ground, which is the part that needs to be filled. In this embodiment, step S300 specifically includes the following steps:

[0080] Step S310: If the target wall column is cylindrical, select the coordinates of the three grounding points where the target wall column contacts the ground from the initial grounding data of the target wall column;

[0081] Step S320: Calculate the coordinates of the center of the circle of the bottom surface of the target wall column in contact with the ground and the radius of the bottom surface based on the coordinates of the three grounding points;

[0082] Step S330: Determine the coordinates of all grounding points of the target wall column in contact with the ground based on the coordinates of the center of the bottom surface and the radius of the bottom surface, thereby obtaining all grounding data of the target wall column.

[0083] Specifically, the equation for all ground contact points of the cylindrical target wall in contact with the ground can be expressed using formula (1):

[0084] (xa) 2 +(yb) 2 =r 2 Formula (1);

[0085] Where x represents the x-coordinate of the grounding point, y represents the y-coordinate of the grounding point, a represents the x-coordinate of the center of the bottom circle, b represents the y-coordinate of the center of the bottom circle, and r represents the radius of the bottom.

[0086] Figure 5 This is a schematic diagram illustrating that, according to one embodiment of the present invention, the visible portion of the target wall column in contact with the ground is shaped as two connected straight lines. Figure 6 Yes Figure 5 A schematic diagram showing the bottom surface after filling where the target wall / column meets the ground. (See diagram below.) Figure 5 and Figure 6 As shown, the solid lines represent the visible portion of the target wall, and the dashed lines represent the invisible portion of the target wall where it meets the ground, which is the part that needs to be filled. In this embodiment, step S300 further includes the following steps:

[0087] Step S340: If the target wall column is a quadrangular prism, then obtain the lengths and coordinates of the two straight lines that connect the target wall column to the ground from the initial grounding data of the target wall column.

[0088] Step S350: Determine the lengths and coordinates of the other two straight lines that connect the target wall column to the ground based on the lengths and coordinates of the two straight lines, thereby obtaining all the grounding data of the target wall column.

[0089] In this embodiment, if the target wall column is a quadrangular prism, then the shape of the target wall in contact with the bottom surface is a parallelogram, including a rectangle. By using the property that opposite sides of a parallelogram are parallel and equal, and knowing the lengths of the visible sides AB and AD of the ground part of the wall column, the coordinates of the positions of point C, BC, CD, and other invisible parts can be calculated.

[0090] In this embodiment, step S300 further includes the following steps:

[0091] Step S360: If the target wall column is a polygonal prism, then obtain the lengths of at least three straight lines that the target wall column contacts the ground, the angles between two adjacent straight lines, and the coordinates of at least three straight lines from the initial grounding data of the target wall column.

[0092] Step S370: Based on the lengths of at least three straight lines in contact with the ground, the angle between two adjacent straight lines, and the coordinates of the at least three straight lines, the lengths and coordinates of the remaining straight lines in contact with the ground are determined, thus obtaining all grounding data for the target wall column. It should be noted that steps S310, S340, and S360 are not sequential.

[0093] For example, if the detected target wall column is a polygonal prism, and it is a pentagonal prism, then the visible portion consists of three straight lines, and the remaining two lines need to be filled in. If it is a hexagonal prism, then the visible portion consists of three straight lines, and the remaining three lines need to be filled in. Specifically, the type of the target wall can be determined by the included angle between adjacent straight lines, such as pentagonal or hexagonal prism.

[0094] In this embodiment, after obtaining all the grounding data of the target wall, it is necessary to combine it with the preset height of the target wall to output the accurate spatial position of the target wall column for use by downstream modules. Here, the preset height can be set to an empirical value; for example, if the height of the wall column is generally 2m, then the preset height is set to 2m.

[0095] In real-world scenarios, parking pillars are typically tall and come in various shapes. In existing technologies, a single image cannot cover the entire pillar, and the area behind the pillar is not visible. Conventional object detection box schemes cannot accurately detect the precise location of pillars with special shapes, such as cylinders. Furthermore, when the vehicle is close to the pillar, the grounding information of the pillar cannot be seen in the image, which significantly affects the accurate location judgment of the pillar and poses a collision risk during vehicle parking.

[0096] This embodiment addresses the aforementioned situation by inputting a bird's-eye view image stitched together from four surround-view fisheye images. After processing by a deep learning detection network using instance segmentation, it outputs the coordinates of the wall / pillar grounding points and the wall / pillar type in the image coordinate system. Using camera intrinsic and extrinsic parameters, the wall / pillar grounding positions in the image coordinate system are transformed to the vehicle coordinate system in three-dimensional space. Considering the geometric type of the wall / pillar, post-processing is used to complete the information of the invisible wall / pillar grounding parts in the image, finally outputting complete and accurate wall / pillar positions. This visual detection scheme can distinguish target wall / pillars and can also be extended to detect other ground obstacles, solving the problem that ultrasonic solutions cannot determine obstacle categories and reducing development costs. Furthermore, this embodiment uses the stitched bird's-eye view image as input, enabling comprehensive detection of wall / pillar information around the vehicle, avoiding the resource consumption of detecting each of the four original images separately. This embodiment can also share the backbone with the parking space detection network, avoiding the resource consumption of adding an additional perception model. Due to visual blind spots caused by the specific shape of the wall pillars and the position of the wall pillars on vehicles, the detection of wall pillars is incomplete and the position is inaccurate. By determining the type of wall pillar and completing the post-processing, the accuracy of wall pillar detection can be effectively improved.

[0097] In this embodiment, step S210 specifically includes the following steps:

[0098] Step S211: Select a feature map from the bird's-eye view;

[0099] Step S212: Use upsampling and downsampling to obtain feature layers of different sizes from the feature map and form a feature pyramid;

[0100] Step S213: Output semantic segmentation results and instance segmentation results for feature layers of different sizes in the feature pyramid, respectively;

[0101] Step S214: Superimpose the semantic segmentation results and instance segmentation results of all feature layers to obtain the location of the target wall column and ground information.

[0102] In step S211, features are extracted using a shared convolutional neural network backbone. Each detection module selects the required feature map based on the features of the element to be detected, so that it can be subsequently input into different detection modules. The detection modules include a wall and pillar detection module, a parking space detection module, an obstacle detection module, and a ground marking detection module. Steps S211 to S214 are all executed by the wall and pillar detection module.

[0103] This embodiment detects target walls and ground by using semantic segmentation and instance segmentation, which can ensure the integrity of the detection of grounding points of wall columns of different sizes and shapes, accurately determine the actual grounding point location of the wall column, avoid the inaccuracy of wall column location detection using two-dimensional images in the prior art, and avoid the problem of inaccurate wall column location detection when the vehicle is too close to the wall column and the image cannot identify the grounding position of the wall column.

[0104] Figure 7 This is a schematic connection block diagram of a wall column detection system for parking according to an embodiment of the present invention. Figure 7 As shown, in a specific embodiment, the wall pillar detection system 100 for parking includes a control module 10. The control module 10 includes a memory 11 and a processor 12. The memory 11 stores a calculation program, which, when executed by the processor, is used to implement the aforementioned wall pillar detection method. The processor 12 can be a central processing unit (CPU), a digital processing unit, etc. The processor 12 sends and receives data via a communication interface. The memory 11 is used to store the program executed by the processor 12. The memory 11 can be any medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer; it can also be a combination of multiple memories 11. The aforementioned calculation program can be downloaded from a computer-readable storage medium to a corresponding computing / processing device or downloaded to a computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). Here, the control module 10 can be a vehicle body controller.

[0105] For the purposes of this embodiment, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of computer-readable media (a non-exhaustive list) include: an electrical connection (electronic device) having one or more wires, a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, a computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory 11.

[0106] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A method for detecting wall columns used in parking, characterized in that, Includes the following steps: Obtain a bird's-eye view of the area around the vehicle while it is parked; The bird's-eye view is inspected using a deep learning detection network to obtain the type of target wall pillars and their initial grounding data. This process includes the following steps: A deep learning detection network is used to detect the bird's-eye view to identify the location of the target wall column and ground information; The initial grounding data of the target wall column is determined based on the location of the target wall column and the ground information, wherein the initial grounding data represents the data of the visible part of the target wall column in contact with the ground; The shape of the visible portion of the target wall column in contact with the ground is determined based on the initial grounding data of the target wall column; If the visible portion of the target wall column in contact with the ground is arc-shaped, then the target wall column is determined to be cylindrical. If the visible portion of the target wall column in contact with the ground consists of two connected straight lines, then the target wall column is determined to be quadrangular prism. If the visible portion of the target wall column in contact with the ground consists of at least three sequentially connected straight lines, then the target wall column is determined to be polygonal prism, and the polygonal prism has more than four sides on its base. After mapping the initial grounding data of the target wall pillar in the bird's-eye view to the three-dimensional coordinate system in the vehicle, the invisible part of the target wall pillar in contact with the ground is filled in according to the type of the target wall pillar and the initial grounding data of the target wall pillar, so as to obtain the complete grounding data of the target wall pillar. Parking is performed based on all grounding data of the target wall column.

2. The wall and column detection method according to claim 1, characterized in that, If the visible portion of the target wall column in contact with the ground consists of two connected straight lines, then determining the type of the target wall column as a quadrangular prism includes the following steps: If the included angle between two connecting straight lines is equal to 90°, then the target wall column is determined to be of the shape of a square prism. If the angle between two straight lines is less than or greater than 90°, then the target wall column is determined to be a non-square prism, specifically a quadrangular prism.

3. The wall and column detection method according to claim 2, characterized in that, The step of filling in the invisible portion of the target wall column in contact with the ground based on the type of the target wall column and the initial grounding data of the target wall column to obtain the complete grounding data of the target wall column specifically includes the following steps: If the target wall column is cylindrical, then the coordinates of the three grounding points where the target wall column contacts the ground are selected from the initial grounding data of the target wall column; The coordinates of the center of the bottom surface of the target wall column in contact with the ground and the radius of the bottom surface are calculated based on the coordinates of the three grounding points. The coordinates of all grounding points of the target wall column in contact with the ground are determined based on the coordinates of the center of the bottom surface and the radius of the bottom surface, thereby obtaining all grounding data of the target wall column.

4. The wall and column detection method according to claim 1, characterized in that, The step of filling in the invisible portion of the target wall column in contact with the ground based on the type of the target wall column and the initial grounding data of the target wall column to obtain the complete grounding data of the target wall column specifically includes the following steps: If the target wall column is a quadrangular prism, then the lengths and coordinates of the two straight lines that connect the target wall column to the ground are obtained from the initial grounding data of the target wall column. Based on the lengths and coordinates of the two straight lines that connect the target wall column to the ground, the lengths and coordinates of the other two straight lines that connect the target wall column to the ground are determined, thereby obtaining all the grounding data of the target wall column.

5. The wall and column detection method according to claim 1, characterized in that, The step of filling in the invisible portion of the target wall column in contact with the ground based on the type of the target wall column and the initial grounding data of the target wall column to obtain the complete grounding data of the target wall column specifically includes the following steps: If the target wall column is a polygonal prism, then the lengths of at least three straight lines in contact with the ground, the angles between two adjacent straight lines, and the coordinates of at least three straight lines are obtained from the initial grounding data of the target wall column. The length and coordinates of the remaining straight lines in contact with the ground are determined based on the lengths of at least three straight lines in contact with the target wall column, the angle between two adjacent straight lines, and the coordinates of the at least three straight lines, thereby obtaining all the grounding data of the target wall column.

6. The wall and column detection method according to claim 1, characterized in that, The step of using a deep learning detection network to detect the bird's-eye view to identify the location of the target wall column and ground information specifically includes the following steps: Select a feature map from the bird's-eye view; Feature layers of different sizes are obtained from the feature map using upsampling and downsampling methods, and then a feature pyramid is formed. In the feature pyramid, feature layers of different sizes output semantic segmentation results and instance segmentation results respectively; The semantic segmentation results and instance segmentation results of all the feature layers are superimposed to obtain the location of the target wall column and the ground information.

7. A wall column detection system for parking, characterized in that, include: A control module, comprising a memory and a processor, wherein the memory stores a calculation program, which, when executed by the processor, is used to implement the wall column detection method according to any one of claims 1-6.

Images