Parking space detection method and related device

Abstract

The present application discloses a parking space detection method and related devices, the method comprising: obtaining a first fisheye stitched image of a target vehicle at the current moment, the target vehicle being an autonomous driving vehicle traveling in a parking lot; inputting the first fisheye stitched image into a neural network model for processing, outputting N groups of parking space data, N parking space key point sets and N parking space types, the N parking space key point sets and the N parking space types respectively corresponding to the N groups of parking space data, N being an integer greater than or equal to 1; obtaining N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data; correcting the N initial parking space information according to the N parking space key point sets and the N parking space types, and obtaining N target parking space information of the target vehicle at the current moment. The use of the embodiments of the present application is helpful to improve the accuracy and efficiency of parking space detection.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a parking space detection method and related devices. Background Art

[0002] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Research in the field of artificial intelligence includes robotics, natural language processing, computer vision, decision-making and reasoning, human-computer interaction, recommendation and search, basic AI theory, etc.

[0003] Autonomous driving is a mainstream application in the field of artificial intelligence. Autonomous driving technology relies on the collaboration of computer vision, radar, monitoring devices, and global positioning systems to enable motor vehicles to achieve autonomous driving without the need for active human operation. Autonomous vehicles use various computing systems to help transport passengers from one location to another. Some autonomous vehicles may require some initial or continuous input from operators (such as navigators, drivers, or passengers). Autonomous vehicles allow operators to switch from manual mode to autonomous driving mode or a mode in between. Since autonomous driving technology does not require humans to drive motor vehicles, it can theoretically effectively avoid human driving errors, reduce the occurrence of traffic accidents, and improve highway transportation efficiency. Therefore, autonomous driving technology is receiving more and more attention.

[0004] At present, the process of parking space detection is: first, multiple key points, entrance edges and parking space edges of the parking space are segmented based on deep learning, and then different parts of the parking space are combined together through prior information, and finally the final parking space is generated through post-processing; or, first, two key points (K points) of the entrance edge of the parking space are detected based on deep learning, and then the two K points are matched in pairs. If certain prior constraints are met, a local image is cropped out through the two K points, and the local image is input into the neural network model for classification. At the same time, the parking space rotation angle is estimated through the local image, and finally the above information is used for post-processing to generate the final parking space. The above two parking space detection methods result in low accuracy and poor efficiency of parking space detection. Summary of the invention

[0005] The embodiments of the present application provide a parking space detection method and related devices for improving the accuracy and efficiency of parking space detection.

[0006] In a first aspect, an embodiment of the present application provides a parking space detection method, comprising:

[0007] Obtaining a first fisheye stitched image of a target vehicle at the current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot;

[0008] Inputting the first fisheye stitching image into the neural network model for processing, outputting N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1;

[0009] Obtaining N initial parking space information of the target vehicle at the current moment according to N groups of parking space data;

[0010] The N initial parking space information is corrected according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

[0011] It can be seen that, compared with post-processing different parts of the parking space combined by prior information to obtain the final parking space, or post-processing the local image cut out based on the two entrance edge key points that meet the prior constraints and the parking space rotation angle estimated by the local image to obtain the final parking space, in the embodiment of the present application, N initial parking space information of the target vehicle at the current moment is obtained based on N groups of parking space data, and the N initial parking space information is corrected based on N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, and the efficiency of parking space detection is greatly improved.

[0012] In some possible embodiments, the N groups of parking space data include N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles, and the N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles correspond to the N groups of parking space data one by one, respectively. According to the N groups of parking space data, the N initial parking space information of the target vehicle at the current moment is obtained, including:

[0013] Obtaining N initial two-dimensional bounding boxes of the target vehicle at the current moment according to the coordinates of the center points of the N parking spaces, the lengths of the N parking spaces, and the widths of the N parking spaces, wherein the coordinates of the center points of the N parking spaces, the lengths of the N parking spaces, and the widths of the N parking spaces correspond one-to-one to the N initial two-dimensional bounding boxes respectively;

[0014] According to the N initial two-dimensional bounding boxes, the N parking space rotation angles and the pre-stored rotation matrix, N target two-dimensional bounding boxes are obtained. The N target two-dimensional bounding boxes are the N initial parking space information of the target vehicle at the current moment. The N initial two-dimensional bounding boxes and the N parking space rotation angles correspond one-to-one to the N target two-dimensional bounding boxes, respectively.

[0015] It can be seen that in this example, firstly, based on the coordinates of the center points of N parking spaces, the lengths of N parking spaces and the widths of N parking spaces, the N initial two-dimensional bounding boxes of the target vehicle at the current moment are obtained, and then based on the N initial two-dimensional bounding boxes, the rotation angles of N parking spaces and the rotation matrix, the N target two-dimensional bounding boxes (i.e., the N initial parking space information of the target vehicle at the current moment) are obtained, and then based on the N parking space key point sets and the N parking space types, the N initial parking space information is corrected to obtain the N target parking space information of the target vehicle at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of the parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, and the efficiency of parking space detection is greatly improved.

[0016] In some possible embodiments, N initial parking space information is corrected according to N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment, including:

[0017] Obtain a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes;

[0018] Obtaining a parking space type corresponding to each target two-dimensional bounding box;

[0019] If the parking space type corresponding to the target two-dimensional bounding box A is a vertical parking space or a parallel parking space, then vertical-parallel parking space post-processing is performed according to the target two-dimensional bounding box A and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, and the target two-dimensional bounding box A is any one of the N target two-dimensional bounding boxes; or,

[0020] If the parking space type corresponding to the target two-dimensional bounding box B is an oblique parking space, the oblique parking space post-processing is performed according to the target two-dimensional bounding box B and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment. The target two-dimensional bounding box B is any one of the N target two-dimensional bounding boxes.

[0021] It can be seen that in this example, based on the parking space type corresponding to the target two-dimensional bounding box, vertical-parallel parking space post-processing or oblique parking space post-processing is performed respectively to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, which greatly improves the efficiency of parking space detection.

[0022] In some possible embodiments, vertical-parallel parking space post-processing is performed according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, including:

[0023] If the target two-dimensional bounding box A matches the two entry edge key points included in the corresponding parking space key point set, then the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment is obtained according to the length of the two entry edge key points and the target two-dimensional bounding box A; or,

[0024] If the target two-dimensional bounding box A matches the three key points included in the corresponding parking space key point set, the rectangle formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0025] If the target two-dimensional bounding box A matches the four key points included in the corresponding parking space key point set, the rectangle formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0026] If the target two-dimensional bounding box A does not match any of the four key points included in its corresponding parking space key point set, the target two-dimensional bounding box A is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment.

[0027] It can be seen that in this example, based on the number of matches between the target two-dimensional bounding box and its corresponding parking space key point set, different methods are used to obtain the vertical or parallel parking space information of the target vehicle corresponding to the target two-dimensional bounding box at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, which greatly improves the efficiency of parking space detection.

[0028] In some possible embodiments, the oblique parking space post-processing is performed according to the target two-dimensional bounding box B and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, including:

[0029] If the target two-dimensional bounding box B matches the two entry edge key points included in the corresponding parking space key point set, then the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment is obtained according to the two entry edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B and the length of the target two-dimensional bounding box B; or,

[0030] If the target two-dimensional bounding box B matches the three key points included in the corresponding parking space key point set, the parallelogram formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment; or,

[0031] If the target two-dimensional bounding box B matches the four key points included in the corresponding parking space key point set, the parallelogram formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

[0032] It can be seen that in this example, based on the number of matches between the target two-dimensional bounding box and its corresponding parking space key point set, different methods are used to obtain the oblique parking space information of the target vehicle corresponding to the target two-dimensional bounding box at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, which greatly improves the efficiency of parking space detection.

[0033] In some possible embodiments, the method further includes:

[0034] Obtaining the driving data of the target vehicle from the current moment to the next moment adjacent to the current moment;

[0035] updating the position of the N target parking space information according to the driving data to obtain the N target parking space information after the position update;

[0036] Obtaining M target parking space information of the target vehicle at the next moment adjacent to the current moment, where M is an integer greater than or equal to 1;

[0037] If the target parking space information C does not match any of the N target parking space information after the position update, the parking space state corresponding to the target parking space information C is set to a newly added parking space, and the target parking space information C is any one of the M target parking space information; or

[0038] If the target parking space information D matches one of the N target parking space information after the position update, the parking space state corresponding to the target parking space information D is set to the original parking space, and the target parking space information D is any one of the M target parking space information.

[0039] It can be seen that in this example, if the target parking space information C does not match any of the N target parking space information after the position update, then the parking space status corresponding to the target parking space information C is set to a newly added parking space; if the target parking space information D matches one of the N target parking space information after the position update, then the parking space status corresponding to the target parking space information D is set to the original parking space, thereby realizing the parking space tracking of the autonomous driving vehicle traveling in the parking lot.

[0040] In a second aspect, an embodiment of the present application provides a parking space detection device, comprising:

[0041] An obtaining unit, used for obtaining a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot;

[0042] The obtaining unit is further used to input the first fisheye stitching image into the neural network model for processing, and output N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, and N is an integer greater than or equal to 1;

[0043] The obtaining unit is further used to obtain N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data;

[0044] The obtaining unit is further used to correct the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

[0045] In a third aspect, the present application provides a parking space detection device, the device comprising a processor, a communication interface and a memory coupled to each other, wherein:

[0046] A processor, used to obtain a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot;

[0047] The processor is further used to input the first fisheye stitching image into the neural network model for processing, and output N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1;

[0048] The processor is further used to obtain N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data;

[0049] The processor is further used to correct the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium storing computer instructions, and a computer program is executed by hardware (such as a processor, etc.) to implement part or all of the steps of any one of the methods performed by the parking space detection device in the embodiments of the present application.

[0051] In a fifth aspect, an embodiment of the present application provides a computer program product. When the computer program product is run on a computer, the computer executes part or all of the steps of the parking space detection method of the above aspects.

[0052] In a sixth aspect, an embodiment of the present application provides a chip system, the chip system including a processor, which is used to support an electronic device to implement part or all of the steps of the parking space detection method of the above aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0054] Figure 1A is a schematic diagram of a first conventional parking space detection provided by an embodiment of the present application;

[0055] Figure 1B is a schematic diagram of a second conventional parking space detection provided by an embodiment of the present application;

[0056] Figure 2A is a flow chart of a parking space detection method provided in an embodiment of the present application;

[0057] Figure 2B is a schematic diagram of the coordinates of the center point of a parking space provided in an embodiment of the present application;

[0058] Figure 2C is a schematic diagram of a parking space rotation angle provided in an embodiment of the present application;

[0059] Figure 2D is a schematic diagram of obtaining initial parking space information of a target vehicle at the current moment provided by an embodiment of the present application;

[0060] Figure 2E is a schematic diagram of a vertical-parallel parking space provided in an embodiment of the present application;

[0061] Figure 2F is a schematic diagram of an oblique parking space provided in an embodiment of the present application;

[0062] Figure 3A It is a flowchart of an automatic parking method provided in an embodiment of the present application;

[0063] Figure 3B is a flowchart of a method for constructing an autonomous driving map provided in an embodiment of the present application;

[0064] Figure 4 It is a block diagram of the functional units of a parking space detection device provided in an embodiment of the present application;

[0065] Figure 5 is a structural schematic diagram of a parking space detection device provided in an embodiment of the present application;

[0066] Figure 6 It is a structural schematic diagram of a chip system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0067] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method part of the embodiments of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0068] See also Figure 1A , Figure 1A 1 is a schematic diagram of the first existing parking space detection provided in an embodiment of the present application. The first parking space detection detects the parking space by a segmentation method. First, multiple key points, entrance edges and parking space edges of the parking space are segmented based on deep learning. Then, the different components of the parking space are combined for post-processing to generate the final parking space.

[0069] In the segmentation scheme, the neural network model outputs discrete parking space components. There is no semantic information connection between the components, and different components need to be combined together through prior information. In order to handle complex and diverse parking scenarios, the post-processing system requirements will be very high, and the robustness and scalability of the entire autonomous valet parking system (AVP) are poor.

[0070] See also Figure 1B , Figure 1B : is a schematic diagram of the second parking space detection provided in an embodiment of the present application. The second parking space detection is to detect the parking space by detection and classification. First, multiple entrance edge key points of the parking space are detected based on deep learning. Then, the detected entrance edge key points are paired with each other. If certain prior constraints are met, a local image is cropped out through the two entrance edge key points. The local image is input into the neural network model for classification. At the same time, the angle of the parking space is estimated through the local image. Finally, the above information is used to generate the final parking space.

[0071] The detection and classification scheme requires the maintenance of two neural networks (detection and classification) at the same time, and the post-processing is also relatively complex, which reduces the robustness of the entire autonomous valet parking system (AVP).

[0072] In order to solve the problem of low accuracy of parking space detection, an embodiment of the present application provides a parking space detection method and related devices.

[0073] See also Figure 2A , Figure 2A A parking space detection method provided in an embodiment of the present application includes steps 201-204, which are as follows:

[0074] 201. A parking space detection device obtains a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot.

[0075] In some possible embodiments, the parking space detection device obtains a first fisheye stitched image of the target vehicle at the current moment, including:

[0076] The parking space detection device sends multiple fisheye image requests to multiple fisheye cameras, each of the multiple fisheye image requests is used to instruct the corresponding fisheye camera to feed back a fisheye image at the current moment, and the multiple fisheye image requests correspond one-to-one to the multiple fisheye cameras; the parking space detection device receives multiple fisheye images at the current moment sent by the multiple fisheye cameras in response to the multiple fisheye image requests; the parking space detection device splices the multiple fisheye images at the current moment to obtain a first fisheye spliced ​​image of the target vehicle at the current moment; or,

[0077] The parking space detection device receives multiple fisheye images at the current moment sent by multiple fisheye cameras, and the multiple fisheye images at the current moment correspond one-to-one to the multiple fisheye cameras; the parking space detection device splices the multiple fisheye images at the current moment to obtain a first fisheye spliced ​​image of the target vehicle at the current moment.

[0078] The multiple fisheye cameras may include a fisheye camera installed on a vehicle logo of the target vehicle, a fisheye camera installed on a rear license plate, a fisheye camera installed on a left rearview mirror, and a fisheye camera installed on a right rearview mirror.

[0079] It should be noted that the above are only two feasible implementations for the parking space detection device to obtain the first fisheye stitched image of the target vehicle at the current moment. The parking space detection device can also obtain the first fisheye stitched image of the target vehicle at the current moment through other feasible implementations, which are not limited here.

[0080] 202. The parking space detection device inputs the first fisheye stitching image into a neural network model for processing, and outputs N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1.

[0081] The neural network model is a trained model, and the neural network model is used to detect one or more groups of parking space data, one or more parking space key point sets, and one or more parking space types included in the fisheye stitching image.

[0082] 203. The parking space detection device obtains N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data.

[0083] In some possible embodiments, the N groups of parking space data include N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles, and the N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles correspond to the N groups of parking space data one by one, respectively. The parking space detection device obtains N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data, including:

[0084] The parking space detection device obtains N initial two-dimensional bounding boxes of the target vehicle at the current moment according to the coordinates of the N parking space center points, the N parking space lengths and the N parking space widths, and the N parking space center point coordinates, the N parking space lengths and the N parking space widths correspond one-to-one to the N initial two-dimensional bounding boxes respectively;

[0085] The parking space detection device obtains N target two-dimensional bounding boxes according to N initial two-dimensional bounding boxes, N parking space rotation angles and a pre-stored rotation matrix. The N target two-dimensional bounding boxes are the N initial parking space information of the target vehicle at the current moment. The N initial two-dimensional bounding boxes and the N parking space rotation angles correspond one-to-one to the N target two-dimensional bounding boxes, respectively.

[0086] The coordinates of the center point of the parking space are the x-axis coordinate value and the y-axis coordinate value of the center point of the parking space in the coordinate system of the target vehicle at the current moment.

[0087] For example, if Figure 2B As shown, Figure 2B : is a schematic diagram of the coordinates of the center point of a parking space provided in an embodiment of the present application. The x-axis coordinate value of the center point of the parking space in the coordinate system of the target vehicle at the current moment is x1, and the y-axis coordinate value of the center point of the parking space in the coordinate system of the target vehicle at the current moment is y1, that is, the coordinates of the center point of the parking space are (x1, y1).

[0088] The parking space rotation angle is the rotation angle of the parking space compared to the target vehicle.

[0089] For example, if Figure 2C As shown, Figure 2C is a schematic diagram of a parking space rotation angle provided in an embodiment of the present application, where the parking space rotation angle is α.

[0090] Optionally, the parking space detection device obtains a target two-dimensional bounding box of the target vehicle at the current moment according to the initial two-dimensional bounding box, the parking space rotation angle and the pre-stored rotation matrix, including:

[0091] The parking space detection device obtains a plurality of first coordinate points according to the initial two-dimensional boundary box;

[0092] The parking space detection device obtains a plurality of second coordinate points according to the plurality of first coordinate points, the parking space rotation angle and the pre-stored rotation matrix, and the plurality of second coordinate points correspond one-to-one to the plurality of first coordinate points;

[0093] The parking space detection device determines the two-dimensional bounding box formed by the plurality of second coordinate points as the target two-dimensional bounding box of the target vehicle at the current moment.

[0094] It should be noted that the above is only one feasible implementation method for the parking space detection device to obtain the target two-dimensional bounding box of the target vehicle at the current moment. The parking space detection device can also obtain the target two-dimensional bounding box of the target vehicle at the current moment through other feasible implementation methods, which will not be described here.

[0095] The rotation matrix can be pre-stored in the parking space detection device, and the rotation matrix is:

[0096]

[0097] x' is the x-axis coordinate value of the coordinate point P1 in the target two-dimensional bounding box of the target vehicle at the current moment, the coordinate point P1 is any one of all the coordinate points included in the target two-dimensional bounding box of the target vehicle at the current moment, y' is the y-axis coordinate value of the coordinate point P1 in the target two-dimensional bounding box of the target vehicle at the current moment, θ is the rotation angle of the parking space corresponding to the target two-dimensional bounding box of the target vehicle at the current moment, x is the x-axis coordinate value of the coordinate point P in the initial two-dimensional bounding box of the target vehicle at the current moment, the coordinate point P is any one of all the coordinate points included in the initial two-dimensional bounding box of the target vehicle at the current moment, y is the y-axis coordinate value of the coordinate point P in the initial two-dimensional bounding box of the target vehicle at the current moment, and the coordinate point P corresponds to the coordinate point P1.

[0098] For example, if Figure 2D As shown, Figure 2D 1 is a schematic diagram of obtaining the initial parking space information of a target vehicle at the current moment provided by an embodiment of the present application. First, the initial two-dimensional bounding box of the target vehicle at the current moment is obtained based on the coordinates of the center point of the parking space, the length of the parking space, and the width of the parking space. Then, the target two-dimensional bounding box of the target vehicle at the current moment is obtained based on the initial two-dimensional bounding box of the current vehicle at the current moment and the parking space rotation angle and rotation matrix.

[0099] It can be seen that in this example, firstly, based on the coordinates of the center points of N parking spaces, the lengths of N parking spaces and the widths of N parking spaces, the N initial two-dimensional bounding boxes of the target vehicle at the current moment are obtained, and then based on the N initial two-dimensional bounding boxes, the rotation angles of N parking spaces and the rotation matrix, the N target two-dimensional bounding boxes (i.e., the N initial parking space information of the target vehicle at the current moment) are obtained, and then based on the N parking space key point sets and the N parking space types, the N initial parking space information is corrected to obtain the N target parking space information of the target vehicle at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of the parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, and the efficiency of parking space detection is greatly improved.

[0100] 204. The parking space detection device corrects the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain N target parking space information of the target vehicle at the current moment.

[0101] It can be seen that, compared with post-processing different parts of the parking space combined by prior information to obtain the final parking space, or post-processing the local image cut out based on the two entrance edge key points that meet the prior constraints and the parking space rotation angle estimated by the local image to obtain the final parking space, in the embodiment of the present application, N initial parking space information of the target vehicle at the current moment is obtained based on N groups of parking space data, and the N initial parking space information is corrected based on N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, and the efficiency of parking space detection is greatly improved.

[0102] In some possible embodiments, the parking space detection device corrects N initial parking space information according to N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment, including:

[0103] The parking space detection device obtains a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes;

[0104] The parking space detection device obtains the parking space type corresponding to each target two-dimensional bounding box;

[0105] If the parking space type corresponding to the target two-dimensional bounding box A is a vertical parking space or a parallel parking space, the parking space detection device performs vertical-parallel parking space post-processing according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, and the target two-dimensional bounding box A is any one of the N target two-dimensional bounding boxes; or,

[0106] If the parking space type corresponding to the target two-dimensional bounding box B is an oblique parking space, the parking space detection device performs oblique parking space post-processing according to the target two-dimensional bounding box B and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment. The target two-dimensional bounding box B is any one of the N target two-dimensional bounding boxes.

[0107] Before the parking space detection device corrects the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment, the N target two-dimensional bounding boxes and the N parking space key point sets are all out of order.

[0108] Optionally, the parking space detection device obtains a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes, including:

[0109] The parking space detection device obtains the center point of a target two-dimensional bounding box E, where the target two-dimensional bounding box E is any one of the N target two-dimensional bounding boxes;

[0110] The parking space detection device obtains N Euclidean distance values ​​between the center point of the target two-dimensional bounding box E and the N parking space key point sets according to the pre-stored Euclidean distance formula, and the N Euclidean distance values ​​correspond to the N parking space key point sets one by one;

[0111] The parking space detection device determines the parking space key point set corresponding to the minimum Euclidean distance value among the N Euclidean distance values ​​as the parking space key point set corresponding to the target two-dimensional bounding box E.

[0112] The Euclidean distance formula may be pre-stored in the parking space detection device, and the Euclidean distance formula is:

[0113]

[0114] s is the Euclidean distance between the center point of the target two-dimensional bounding box E and the parking space key point set, (x', y') is the center point of the target two-dimensional bounding box E, (a1, b1) is the first key point included in the parking space key point set, (a2, b2) is the second key point included in the parking space key point set, (a3, b3) is the third key point included in the parking space key point set, and (a4, b4) is the fourth key point included in the parking space key point set.

[0115] The parking space key point set may include two entrance edge key points and two parking space edge key points, the entrance edge is the edge for the vehicle to enter the parking space, and the parking space edge is three edges of the four edges of the parking space except the entrance edge.

[0116] It should be noted that the above is only one feasible implementation method for the parking space detection device to obtain the parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes. The parking space detection device can also obtain the parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes through other feasible implementation methods, which is not limited here.

[0117] Parking space types can include perpendicular parking spaces, parallel parking spaces, and diagonal parking spaces.

[0118] It can be seen that in this example, based on the parking space type corresponding to the target two-dimensional bounding box, vertical-parallel parking space post-processing or oblique parking space post-processing is performed respectively to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, which greatly improves the efficiency of parking space detection.

[0119] In some possible embodiments, the parking space detection device performs vertical-parallel parking space post-processing according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, including:

[0120] If the target two-dimensional bounding box A matches the two entry edge key points included in the corresponding parking space key point set, the parking space detection device obtains the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment according to the two entry edge key points and the length of the target two-dimensional bounding box A; or,

[0121] If the target two-dimensional bounding box A matches the three key points included in the corresponding parking space key point set, the parking space detection device determines the rectangle formed by the three key points as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0122] If the target two-dimensional bounding box A matches the four key points included in the corresponding parking space key point set, the parking space detection device determines the rectangle formed by the four key points as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0123] If the target two-dimensional bounding box A does not match any of the four key points included in its corresponding parking space key point set, the parking space detection device determines the target two-dimensional bounding box A as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment.

[0124] Optionally, the parking space detection device obtains target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment according to the two entry edge key points and the length of the target two-dimensional bounding box A, including:

[0125] The parking space detection device connects the two entrance edge key points to obtain a first entrance edge to be connected;

[0126] The parking space detection device uses the two entrance edge key points as endpoints, draws a line segment perpendicular to the entrance edge to be connected and having a length equal to the length of the target two-dimensional bounding box A, and obtains two first parking space edges to be connected;

[0127] The parking space detection device determines the rectangle formed by the first entrance edge to be connected and the two first parking space edges to be connected as the target parking space information of the target vehicle corresponding to the target two-dimensional boundary box A at the current moment.

[0128] For example, if Figure 2E As shown, Figure 2E It is a schematic diagram of a vertical-parallel parking space provided in an embodiment of the present application, wherein the vertical-parallel parking space includes an entrance edge key point K1 and an entrance edge key point K2, two parking space edges included in the vertical-parallel parking space are perpendicular to the entrance edge, another parking space edge included in the vertical-parallel parking space is parallel to the entrance edge, and the length of the vertical-parallel parking space is the same as the length of the target two-dimensional bounding box A.

[0129] It should be noted that the above is only a feasible implementation method for the parking space detection device to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment. The parking space detection device can also obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment through other feasible implementation methods, which are not limited here.

[0130] It can be seen that in this example, based on the number of matches between the target two-dimensional bounding box and its corresponding parking space key point set, different methods are used to obtain the vertical or parallel parking space information of the target vehicle corresponding to the target two-dimensional bounding box at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, which greatly improves the efficiency of parking space detection.

[0131] In some possible embodiments, the parking space detection device performs oblique parking space post-processing according to the target two-dimensional bounding box B and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, including:

[0132] If the target two-dimensional bounding box B matches the two entry edge key points included in the corresponding parking space key point set, the parking space detection device obtains the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment according to the two entry edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B and the length of the target two-dimensional bounding box B; or,

[0133] If the target two-dimensional bounding box B matches the three key points included in the corresponding parking space key point set, the parking space detection device determines the parallelogram formed by the three key points as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment; or,

[0134] If the target two-dimensional bounding box B matches the four key points included in its corresponding parking space key point set, the parking space detection device determines the parallelogram formed by the four key points as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

[0135] Optionally, the parking space detection device obtains target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment according to the two entrance edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B, and the length of the target two-dimensional bounding box B, including:

[0136] The parking space detection device connects the two entrance edge key points to obtain a second entrance edge to be connected;

[0137] The parking space detection device uses the two entry edge key points as endpoints, and constructs a line segment with an angle equal to the parking space rotation angle corresponding to the target two-dimensional bounding box B and a length equal to the length of the target two-dimensional bounding box B, to obtain two second parking space edges to be connected;

[0138] The parking space detection device determines the parallelogram formed by the second entrance edge to be connected and the two second parking space edges to be connected as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

[0139] For example, if Figure 2F As shown, Figure 2F is a schematic diagram of an oblique parking space provided in an embodiment of the present application, the oblique parking space includes an entrance edge key point K1 and an entrance edge key point K2, the angle between any one of the two parking space edges included in the oblique parking space and the entrance edge is the parking space rotation angle α corresponding to the target two-dimensional bounding box B, the other parking space edge included in the oblique parking space is parallel to the entrance edge, and the length of the oblique parking space is the same as the length of the target two-dimensional bounding box B.

[0140] It should be noted that the above is only a feasible implementation method for the parking space detection device to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment. The parking space detection device can also obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment through other feasible implementation methods, which are not limited here.

[0141] Furthermore, the method also includes:

[0142] If the target two-dimensional bounding box B does not match any of the four key points included in its corresponding parking space key point set, the parking space detection device does not perform any operation.

[0143] The parking space detection device does not perform any operation. In other words, the parking space detection device determines that the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment is empty.

[0144] It can be seen that in this example, based on the number of matches between the target two-dimensional bounding box and its corresponding parking space key point set, different methods are used to obtain the oblique parking space information of the target vehicle corresponding to the target two-dimensional bounding box at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, which greatly improves the efficiency of parking space detection.

[0145] In some possible embodiments, the method further includes:

[0146] The parking space detection device obtains the driving data of the target vehicle from the current moment to the next moment adjacent to the current moment;

[0147] The parking space detection device updates the position of the N target parking space information according to the driving data to obtain the N target parking space information after the position update;

[0148] The parking space detection device obtains M target parking space information of the target vehicle at the next moment adjacent to the current moment, where M is an integer greater than or equal to 1;

[0149] If the target parking space information C does not match any of the N target parking space information after the position update, the parking space detection device sets the parking space state corresponding to the target parking space information C to a newly added parking space, and the target parking space information C is any one of the M target parking space information; or

[0150] If the target parking space information D matches one of the N target parking space information after the position update, the parking space detection device sets the parking space state corresponding to the target parking space information D to the original parking space, and the target parking space information D is any one of the M target parking space information.

[0151] The driving data may include driving distance.

[0152] Optionally, the parking space detection device updates the target parking space information according to the driving data to obtain the updated target parking space information, including:

[0153] The parking space detection device obtains the position of the parking space F in the coordinate system of the target vehicle at the current moment according to the target parking space information;

[0154] The parking space detection device obtains the position of the parking space F in the coordinate system of the target vehicle at a next moment adjacent to the current moment according to the position of the parking space F in the coordinate system of the target vehicle at the current moment and the travel distance;

[0155] The parking space detection device determines the position of the parking space F in the coordinate system of the target vehicle at the next time point adjacent to the current time point as the target parking space information after the position update.

[0156] When the target vehicle is traveling in a straight line, if the position of parking space F in the coordinate system of the target vehicle at the current moment includes (a5, b5), (a6, b6), (a7, b7) and (a8, b8), and the driving distance is d, then the position of parking space F in the coordinate system of the target vehicle at the next moment adjacent to the current moment includes (a5-d+d', b5-d+d'), (a6-d+d', b6-d+d'), (a7-d+d', b7-d+d') and (a8-d+d', b8-d+d'); wherein d' may be preset, in other words, professionals may manually set d' based on historical measurement data.

[0157] It should be noted that the above is only one feasible implementation method for the parking space detection device to obtain the parking space information after the location is updated. The parking space detection device can also obtain the target parking space information after the location is updated through other feasible implementation methods, which are not limited here.

[0158] The target parking space information C does not match any of the N target parking space information after the position update. In other words, the parking space included in the target parking space information C is newly detected by the parking space detection device.

[0159] The target parking space information C matches one of the N target parking space information after the position update. In other words, the parking space included in the target parking space information C is previously detected by the parking space detection device.

[0160] Furthermore, the method also includes:

[0161] During the process of the parking space detection device tracking the parking space, if the parking space G is not detected within a preset time period, the parking space detection device deletes the parking space G, and the parking space G is the parking space detected by the parking space detection device.

[0162] The preset duration may be pre-set, for example, the preset duration may be three times the difference between the next moment adjacent to the current moment and the current moment.

[0163] It can be seen that in this example, if the target parking space information C does not match any of the N target parking space information after the position update, then the parking space status corresponding to the target parking space information C is set to a newly added parking space; if the target parking space information D matches one of the N target parking space information after the position update, then the parking space status corresponding to the target parking space information D is set to the original parking space, thus realizing the parking space tracking of the autonomous driving vehicle driving in the parking lot. In addition, the parking spaces that have not been detected within the preset time are deleted, so as to control the number of parking spaces tracked to be kept within a reasonable range, thereby improving the efficiency of the parking space tracking of the autonomous driving vehicle driving in the parking lot.

[0164] See also Figure 3A , Figure 3A An automatic parking method is provided in an embodiment of the present application. The automatic parking method includes steps 301-310, which are as follows:

[0165] 301. The perception module obtains a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot.

[0166] The description of step 301 can be found in the above step 201 and will not be described again here.

[0167] 302. The perception module inputs the first fisheye stitched image into the neural network model for processing, and outputs N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1.

[0168] The description of step 302 can be found in the above step 202 and will not be described again here.

[0169] 303. The perception module obtains N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data.

[0170] For step 303, please refer to the description of step 203 above, which will not be described again here.

[0171] 304. The perception module corrects the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

[0172] The parking space types may include unoccupied perpendicular parking space, occupied perpendicular parking space, unoccupied parallel parking space, occupied parallel parking space, unoccupied diagonal parking space, and occupied diagonal parking space.

[0173] For step 304, please refer to the description of step 204 above, which will not be described again here.

[0174] 305. The perception module sends the information of the N target parking spaces to the data fusion center.

[0175] 306. The planning and decision module sends a parking space information request to the data fusion center. The parking space information request is used to instruct the data fusion center to feed back one or more parking space information stored therein.

[0176] 307. The data fusion center receives the parking space information request sent by the planning and decision module, and sends N target parking space information to the decision and planning module.

[0177] 308. The decision-making planning module receives the N target parking space information sent by the data fusion center, and obtains the vehicle driving route for the target vehicle to automatically park into the garage according to the N target parking space information.

[0178] Optionally, the decision-making planning module obtains a vehicle driving route for the target vehicle to automatically park into the garage according to the information of the N target parking spaces, including:

[0179] The decision-making and planning module obtains one or more unoccupied parking spaces and one or more parking space positions according to the N target parking space information, and the one or more parking space positions correspond to the one or more unoccupied parking spaces one by one;

[0180] The decision-making planning module determines any one of the one or more unoccupied parking spaces as a parking space for the target vehicle to automatically park into the garage;

[0181] The decision-making planning module obtains a vehicle driving route for the target vehicle to automatically park into the garage according to the parking space position corresponding to the parking space where the target vehicle automatically parks into the garage and a pre-stored path planning algorithm.

[0182] It should be noted that the above is only a feasible implementation method for the decision-making planning module to obtain the vehicle driving route of the target vehicle for automatic parking. The decision-making planning module can also obtain the vehicle driving route of the target vehicle for automatic parking through other feasible implementation methods, which is not limited here.

[0183] 309. The decision-making and planning module sends the vehicle driving route to the control module.

[0184] 310. The control module receives the vehicle driving route sent by the decision-making and planning module, and performs an automatic parking operation according to the vehicle driving route.

[0185] It can be seen that in the embodiment of the present application, N initial parking space information of the target vehicle at the current moment is obtained based on N groups of parking space data, and the N initial parking space information is corrected based on N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, and the efficiency of parking space detection is greatly improved. The decision-making planning module obtains the vehicle driving route of the target vehicle for automatic parking based on the N target parking space information, so that the control module performs the automatic parking operation based on the vehicle driving route, thereby improving the perception accuracy of the parking space in the process of assisting automatic valet parking.

[0186] See also Figure 3B , Figure 3B The present invention provides a method for constructing an autonomous driving map. The method for constructing an autonomous driving map includes steps 311-318, which are as follows:

[0187] 311. The perception module obtains a first fisheye stitched image of the target vehicle at the current moment, where the target vehicle is an autonomous driving vehicle traveling in the parking lot.

[0188] The description of step 311 can be found in the above step 201 and will not be described again here.

[0189] 312. The perception module inputs the first fisheye stitched image into the neural network model for processing, and outputs N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1.

[0190] The description of step 312 can be found in the above step 202 and will not be described again here.

[0191] 313. The perception module obtains N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data.

[0192] For step 313, please refer to the description of step 203 above, which will not be described again here.

[0193] 314. The perception module corrects the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

[0194] For step 314, please refer to the description of step 204 above, which will not be described again here.

[0195] 315. The perception module sends the information of the N target parking spaces to the data fusion center.

[0196] 316. The map module sends a parking space information request to the data fusion center. The parking space information request is used to instruct the data fusion center to feedback one or more parking space information stored therein.

[0197] 317. The data fusion center receives the parking space information request sent by the map module, and sends N target parking space information to the map module.

[0198] 318. The map module receives the N target parking space information sent by the data fusion center, and constructs an automatic driving map according to the N target parking space information.

[0199] Optionally, the map module constructs an autonomous driving map according to the information of N target parking spaces, including:

[0200] The map module obtains the position of each of the N parking spaces included in the N target parking space information in a pre-stored parking lot coordinate system according to the N target parking space information;

[0201] The map module constructs an autonomous driving map according to the position of each of the N parking spaces in the parking lot coordinate system and a pre-stored parking lot map.

[0202] It should be noted that the above is only one feasible implementation method for the map module to construct an autonomous driving map. The map module can also construct an autonomous driving map through other feasible implementation methods, which are not limited here.

[0203] It can be seen that in the embodiment of the present application, N initial parking space information of the target vehicle at the current moment is obtained based on N groups of parking space data, and the N initial parking space information is corrected based on N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment. Since the neural network model outputs the key information and semantic information of the parking space, it helps to improve the accuracy of parking space detection. In addition, in the process of generating one or more target parking space information based on the key information and semantic information of the parking space, the complexity of post-processing is reduced, and the efficiency of parking space detection is greatly improved. The map module constructs an autonomous driving map based on the N target parking space information, thereby improving the efficiency of drawing maps in the process of assisted automatic valet parking.

[0204] See also Figure 4 , Figure 4 4 is a functional unit block diagram of a parking space detection device provided in an embodiment of the present application. The parking space detection device 400 includes an obtaining unit 410, wherein:

[0205] An obtaining unit 410 is used to obtain a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot;

[0206] The obtaining unit 410 is further configured to input the first fisheye stitched image into the neural network model for processing, and output N groups of parking space data, N parking space key point sets, and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, and N is an integer greater than or equal to 1;

[0207] The obtaining unit 410 is further used to obtain N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data;

[0208] The obtaining unit 410 is further configured to correct the N initial parking space information according to the N parking space key point sets and the N parking space types, so as to obtain the N target parking space information of the target vehicle at the current moment.

[0209] In some possible implementations, the N groups of parking space data include N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles, and the N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles correspond to the N groups of parking space data one by one, respectively. In terms of obtaining N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data, the obtaining unit 410 is specifically used to:

[0210] Obtaining N initial two-dimensional bounding boxes of the target vehicle at the current moment according to the coordinates of the center points of the N parking spaces, the lengths of the N parking spaces, and the widths of the N parking spaces, wherein the coordinates of the center points of the N parking spaces, the lengths of the N parking spaces, and the widths of the N parking spaces correspond one-to-one to the N initial two-dimensional bounding boxes respectively;

[0211] According to the N initial two-dimensional bounding boxes, the N parking space rotation angles and the pre-stored rotation matrix, N target two-dimensional bounding boxes are obtained. The N target two-dimensional bounding boxes are the N initial parking space information of the target vehicle at the current moment. The N initial two-dimensional bounding boxes and the N parking space rotation angles correspond one-to-one to the N target two-dimensional bounding boxes, respectively.

[0212] In some possible implementations, in terms of correcting N initial parking space information according to N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment, the obtaining unit 410 is specifically configured to:

[0213] Obtain a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes;

[0214] Obtaining a parking space type corresponding to each target two-dimensional bounding box;

[0215] If the parking space type corresponding to the target two-dimensional bounding box A is a vertical parking space or a parallel parking space, then vertical-parallel parking space post-processing is performed according to the target two-dimensional bounding box A and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, and the target two-dimensional bounding box A is any one of the N target two-dimensional bounding boxes; or,

[0216] If the parking space type corresponding to the target two-dimensional bounding box B is an oblique parking space, the oblique parking space post-processing is performed according to the target two-dimensional bounding box B and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment. The target two-dimensional bounding box B is any one of the N target two-dimensional bounding boxes.

[0217] In some possible implementations, in terms of performing vertical-parallel parking space post-processing according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, the obtaining unit 410 is specifically used to:

[0218] If the target two-dimensional bounding box A matches the two entry edge key points included in the corresponding parking space key point set, then the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment is obtained according to the length of the two entry edge key points and the target two-dimensional bounding box A; or,

[0219] If the target two-dimensional bounding box A matches the three key points included in the corresponding parking space key point set, the rectangle formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0220] If the target two-dimensional bounding box A matches the four key points included in the corresponding parking space key point set, the rectangle formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0221] If the target two-dimensional bounding box A does not match any of the four key points included in its corresponding parking space key point set, the target two-dimensional bounding box A is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment.

[0222] In some possible implementations, in terms of performing oblique parking space post-processing according to the target two-dimensional bounding box B and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, the obtaining unit 410 is specifically used to:

[0223] If the target two-dimensional bounding box B matches the two entry edge key points included in the corresponding parking space key point set, then the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment is obtained according to the two entry edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B and the length of the target two-dimensional bounding box B; or,

[0224] If the target two-dimensional bounding box B matches the three key points included in the corresponding parking space key point set, the parallelogram formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment; or,

[0225] If the target two-dimensional bounding box B matches the four key points included in the corresponding parking space key point set, the parallelogram formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

[0226] In some possible implementations,

[0227] The obtaining unit 410 is further used to obtain the driving data of the target vehicle from the current moment to the next moment adjacent to the current moment;

[0228] The obtaining unit 410 is further used to update the position of the N target parking space information according to the driving data to obtain the N target parking space information after the position is updated;

[0229] The obtaining unit 410 is further used to obtain M target parking space information of the target vehicle at the next moment adjacent to the current moment, where M is an integer greater than or equal to 1;

[0230] The obtaining unit 410 is further configured to set the parking space state corresponding to the target parking space information C to a newly added parking space, and the target parking space information C is any one of the M target parking space information, if the target parking space information C does not match any of the N target parking space information after the position is updated; or, if the target parking space information D matches any one of the N target parking space information after the position is updated, set the parking space state corresponding to the target parking space information D to an original parking space, and the target parking space information D is any one of the M target parking space information.

[0231] See also Figure 5 , Figure 5 is a schematic diagram of the structure of a parking space detection device provided in an embodiment of the present application. The parking space detection device 500 includes a memory 510 , a communication interface 520 , and a processor 530 coupled to each other; for example, the memory 510 , the communication interface 520 , and the processor 530 are coupled via a bus 540 .

[0232] The memory 510 may include but is not limited to random access memory (Random Access Memory, RAM), erasable programmable ROM (Erasable Programmable ROM, EPROM), read-only memory (Read-Only Memory, ROM) or portable read-only memory (Compact Disc Read-Only Memory, CD-ROM), etc. The memory 510 is used for related instructions and data.

[0233] The processor 530 may be one or more central processing units (CPUs). When the processor 530 is a CPU, the CPU may be a single-core CPU or a multi-core CPU.

[0234] The processor 530 is used to read the program code stored in the memory 510 and cooperate with the communication interface 520 to execute part or all of the steps of the method performed by the parking space detection device 500 in the above embodiment of the present application.

[0235] Processor 530, used to obtain a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot;

[0236] The processor 530 is further configured to input the first fisheye stitched image into the neural network model for processing, and output N groups of parking space data, N parking space key point sets, and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, and N is an integer greater than or equal to 1;

[0237] The processor 530 is further configured to obtain N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data;

[0238] The processor 530 is further configured to correct the N initial parking space information according to the N parking space key point sets and the N parking space types, so as to obtain the N target parking space information of the target vehicle at the current moment.

[0239] In some possible embodiments, the N sets of parking space data include N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles, and the N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles correspond to the N sets of parking space data one by one, respectively. In terms of obtaining N initial parking space information of the target vehicle at the current moment according to the N sets of parking space data, the processor 530 is specifically configured to:

[0240] Obtaining N initial two-dimensional bounding boxes of the target vehicle at the current moment according to the coordinates of the center points of the N parking spaces, the lengths of the N parking spaces, and the widths of the N parking spaces, wherein the coordinates of the center points of the N parking spaces, the lengths of the N parking spaces, and the widths of the N parking spaces correspond one-to-one to the N initial two-dimensional bounding boxes respectively;

[0241] According to the N initial two-dimensional bounding boxes, the N parking space rotation angles and the pre-stored rotation matrix, N target two-dimensional bounding boxes are obtained. The N target two-dimensional bounding boxes are the N initial parking space information of the target vehicle at the current moment. The N initial two-dimensional bounding boxes and the N parking space rotation angles correspond one-to-one to the N target two-dimensional bounding boxes, respectively.

[0242] In some possible embodiments, in terms of correcting N initial parking space information according to N parking space key point sets and N parking space types to obtain N target parking space information of the target vehicle at the current moment, the processor 530 is specifically configured to:

[0243] Obtain a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes;

[0244] Obtaining a parking space type corresponding to each target two-dimensional bounding box;

[0245] If the parking space type corresponding to the target two-dimensional bounding box A is a vertical parking space or a parallel parking space, then vertical-parallel parking space post-processing is performed according to the target two-dimensional bounding box A and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, and the target two-dimensional bounding box A is any one of the N target two-dimensional bounding boxes; or,

[0246] If the parking space type corresponding to the target two-dimensional bounding box B is an oblique parking space, the oblique parking space post-processing is performed according to the target two-dimensional bounding box B and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment. The target two-dimensional bounding box B is any one of the N target two-dimensional bounding boxes.

[0247] In some possible embodiments, in performing vertical-parallel parking space post-processing according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, the processor 530 is specifically configured to:

[0248] If the target two-dimensional bounding box A matches the two entry edge key points included in the corresponding parking space key point set, then the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment is obtained according to the length of the two entry edge key points and the target two-dimensional bounding box A; or,

[0249] If the target two-dimensional bounding box A matches the three key points included in the corresponding parking space key point set, the rectangle formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0250] If the target two-dimensional bounding box A matches the four key points included in the corresponding parking space key point set, the rectangle formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or,

[0251] If the target two-dimensional bounding box A does not match any of the four key points included in its corresponding parking space key point set, the target two-dimensional bounding box A is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment.

[0252] In some possible embodiments, in performing oblique parking space post-processing according to the target two-dimensional bounding box B and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, the processor 530 is specifically configured to:

[0253] If the target two-dimensional bounding box B matches the two entry edge key points included in the corresponding parking space key point set, then the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment is obtained according to the two entry edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B and the length of the target two-dimensional bounding box B; or,

[0254] If the target two-dimensional bounding box B matches the three key points included in the corresponding parking space key point set, the parallelogram formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment; or,

[0255] If the target two-dimensional bounding box B matches the four key points included in the corresponding parking space key point set, the parallelogram formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

[0256] In some possible embodiments,

[0257] The processor 530 is further used to obtain the driving data of the target vehicle from the current moment to the next moment adjacent to the current moment;

[0258] The processor 530 is further configured to update the position of the N target parking space information according to the driving data to obtain the N target parking space information after the position is updated;

[0259] The processor 530 is further configured to obtain M target parking space information of the target vehicle at a next moment adjacent to the current moment, where M is an integer greater than or equal to 1;

[0260] The processor 530 is further configured to set the parking space state corresponding to the target parking space information C to a newly added parking space if the target parking space information C does not match any of the N target parking space information after the position is updated, and the target parking space information C is any one of the M target parking space information; or, if the target parking space information D matches one of the N target parking space information after the position is updated, set the parking space state corresponding to the target parking space information D to an existing parking space, and the target parking space information D is any one of the M target parking space information.

[0261] The embodiment of the present application also provides a computer-readable storage medium storing computer instructions, and the computer program is executed by hardware (such as a processor, etc.) to implement part or all of the steps of any method performed by the parking space detection device in the embodiment of the present application.

[0262] The embodiments of the present application further provide a computer program product. When the computer program product is executed on a computer or a processor, the computer or the processor executes part or all of the steps of the parking space detection method in the above aspects.

[0263] See also Figure 6 , Figure 6 6 is a schematic diagram of a chip system according to an embodiment of the present application. The chip system 600 may include: a processor 610, and one or more interfaces 620 coupled to the processor 610. Exemplary:

[0264] The processor 610 can be used to read and execute computer-readable instructions. In a specific implementation, the processor 610 may mainly include a controller, an arithmetic unit and a register. Exemplarily, the controller is mainly responsible for decoding instructions and issuing control signals for operations corresponding to the instructions. The arithmetic unit is mainly responsible for performing fixed-point or floating-point arithmetic operations, shift operations, and logical operations, etc., and may also perform address operations and conversions. The register is mainly responsible for storing register operands and intermediate operation results temporarily stored during the execution of instructions. In a specific implementation, the hardware architecture of the processor 610 may be an application specific integrated circuit (ASIC) architecture, a microprocessor without interlocked piped stages architecture (MIPS) architecture, an advanced RISC machines (ARM) architecture or an NP architecture, etc. The processor 610 may be single-core or multi-core.

[0265] Exemplarily, the interface 620 can be used to input data to be processed to the processor 610, and can output the processing results of the processor 610. In a specific implementation, the interface 620 can be a general purpose input output (GPIO) interface, which can be connected to multiple peripheral devices (such as a display (LCD), a camera (camara), a radio frequency (RF) module, etc.). The interface 620 is connected to the processor 610 via a bus 630.

[0266] In some possible embodiments, the processor 610 may be used to call from the memory the implementation program or data of the parking space detection method, the automatic parking method, and the method for constructing an automatic driving map provided in one or more embodiments of the present application on the network device or terminal device side, so that the chip can implement the aforementioned Figure 2A The parking space detection method shown, Figure 3A The automatic parking method and Figure 3B The method for constructing an autonomous driving map shown. The memory can be integrated with the processor 610, or it can be coupled to the chip system 600 through the interface 620, that is, the memory can be a part of the chip system 600, or it can be independent of the chip system 600. The interface 620 can be used to output the execution result of the processor 610. In the present application, the interface 620 can be specifically used to output the decoding result of the processor 610. Regarding the parking space detection method provided in one or more embodiments of the present application, reference can be made to the aforementioned embodiments, which will not be repeated here.

[0267] In the above embodiments, all or part of the functions can be implemented by software, hardware, or a combination of software and hardware. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrations. Available media can be magnetic media, (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid state drives (SSDs)), etc.

[0268] Those skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by computer programs to instruct related hardware, and the programs can be stored in computer-readable storage media. When the programs are executed, they can include the processes of the above-mentioned method embodiments. The aforementioned storage media include: ROM or random access memory RAM, magnetic disk or optical disk and other media that can store program codes.

Claims

1. A parking space detection method, characterized in that: include: Obtaining a first fisheye stitched image of a target vehicle at a current moment, wherein the target vehicle is an autonomous driving vehicle traveling in a parking lot; Inputting the first fisheye stitched image into a neural network model for processing, and outputting N groups of parking space data, N parking space key point sets, and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1; Obtaining N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data; The N initial parking space information is corrected according to the N parking space key point sets and the N parking space types to obtain N target parking space information of the target vehicle at the current moment.

2. The method according to claim 1, characterized in that The N groups of parking space data include N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles, and the N parking space center point coordinates, the N parking space lengths, the N parking space widths, and the N parking space rotation angles correspond to the N groups of parking space data one by one, respectively. The obtaining of the N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data includes: Obtaining N initial two-dimensional bounding boxes of the target vehicle at the current moment according to the N parking space center point coordinates, the N parking space lengths, and the N parking space widths, wherein the N parking space center point coordinates, the N parking space lengths, and the N parking space widths correspond one-to-one to the N initial two-dimensional bounding boxes respectively; N target two-dimensional bounding boxes are obtained according to the N initial two-dimensional bounding boxes, the N parking space rotation angles and the pre-stored rotation matrix, the N target two-dimensional bounding boxes are the N initial parking space information of the target vehicle at the current moment, and the N initial two-dimensional bounding boxes and the N parking space rotation angles correspond one-to-one to the N target two-dimensional bounding boxes respectively.

3. The method according to claim 2, characterized in that The correcting the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment includes: Obtain a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes; Obtaining a parking space type corresponding to each target two-dimensional bounding box; If the parking space type corresponding to the target two-dimensional bounding box A is a vertical parking space or a parallel parking space, then vertical-parallel parking space post-processing is performed according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, and the target two-dimensional bounding box A is any one of the N target two-dimensional bounding boxes; or, If the parking space type corresponding to the target two-dimensional bounding box B is an oblique parking space, the oblique parking space post-processing is performed according to the target two-dimensional bounding box B and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, and the target two-dimensional bounding box B is any one of the N target two-dimensional bounding boxes.

4. The method according to claim 3, characterized in that The performing vertical-parallel parking space post-processing according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment includes: If the target two-dimensional bounding box A matches two entry edge key points included in its corresponding parking space key point set, then obtaining the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment according to the two entry edge key points and the length of the target two-dimensional bounding box A; or If the target two-dimensional bounding box A matches the three key points included in the corresponding parking space key point set, a rectangle formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or If the target two-dimensional bounding box A matches the four key points included in the corresponding parking space key point set, the rectangle formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or If the target two-dimensional bounding box A does not match any of the four key points included in its corresponding parking space key point set, the target two-dimensional bounding box A is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment.

5. The method according to claim 3, characterized in that: The step of performing oblique parking space post-processing according to the target two-dimensional bounding box B and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment includes: If the target two-dimensional bounding box B matches two entry edge key points included in its corresponding parking space key point set, then obtaining the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment according to the two entry edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B and the length of the target two-dimensional bounding box B; or, If the target two-dimensional bounding box B matches the three key points included in the corresponding parking space key point set, the parallelogram formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment; or If the target two-dimensional bounding box B matches the four key points included in the corresponding parking space key point set, the parallelogram formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

6. The method according to claim 4 or 5, characterized in that: The method further comprises: Obtaining driving data of the target vehicle from the current moment to a next moment adjacent to the current moment; updating the positions of the N target parking space information according to the driving data to obtain the N target parking space information after position update; Obtaining M target parking space information of the target vehicle at the next moment, where M is an integer greater than or equal to 1; If the target parking space information C does not match the N target parking space information after the position update, the parking space state corresponding to the target parking space information C is set to a newly added parking space, and the target parking space information C is any one of the M target parking space information; or If the target parking space information D matches one of the N target parking space information after the position update, the parking space state corresponding to the target parking space information D is set to the original parking space, and the target parking space information D is any one of the M target parking space information.

7. A parking space detection device, characterized in that: include: An obtaining unit, configured to obtain a first fisheye stitched image of a target vehicle at a current moment, wherein the target vehicle is an autonomous driving vehicle traveling in a parking lot; The obtaining unit is further used for inputting the first fisheye stitching image into a neural network model for processing, and outputting N groups of parking space data, N parking space key point sets and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1; The obtaining unit is further used to obtain N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data; The obtaining unit is further used to correct the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

8. The device according to claim 7, characterized in that The N groups of parking space data include N parking space center point coordinates, N parking space lengths, N parking space widths, and N parking space rotation angles, and the N parking space center point coordinates, the N parking space lengths, the N parking space widths, and the N parking space rotation angles correspond to the N groups of parking space data one by one, respectively. In terms of obtaining the N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data, the obtaining unit is specifically used to: Obtaining N initial two-dimensional bounding boxes of the target vehicle at the current moment according to the N parking space center point coordinates, the N parking space lengths, and the N parking space widths, wherein the N parking space center point coordinates, the N parking space lengths, and the N parking space widths correspond one-to-one to the N initial two-dimensional bounding boxes respectively; N target two-dimensional bounding boxes are obtained according to the N initial two-dimensional bounding boxes, the N parking space rotation angles and the pre-stored rotation matrix, wherein the N target two-dimensional bounding boxes are the N initial parking space information of the target vehicle at the current moment. The N initial two-dimensional bounding boxes and the N parking space rotation angles correspond one-to-one to the N target two-dimensional bounding boxes, respectively.

9. The device according to claim 8, characterized in that In terms of correcting the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment, the obtaining unit is specifically used to: Obtain a parking space key point set corresponding to each target two-dimensional bounding box in the N target two-dimensional bounding boxes; Obtaining a parking space type corresponding to each target two-dimensional bounding box; If the parking space type corresponding to the target two-dimensional bounding box A is a vertical parking space or a parallel parking space, then vertical-parallel parking space post-processing is performed according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, and the target two-dimensional bounding box A is any one of the N target two-dimensional bounding boxes; or, If the parking space type corresponding to the target two-dimensional bounding box B is an oblique parking space, the oblique parking space post-processing is performed according to the target two-dimensional bounding box B and its corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, and the target two-dimensional bounding box B is any one of the N target two-dimensional bounding boxes.

10. The device according to claim 9, characterized in that In terms of performing vertical-parallel parking space post-processing according to the target two-dimensional bounding box A and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment, the obtaining unit is specifically used to: If the target two-dimensional bounding box A matches two entry edge key points included in the corresponding parking space key point set, then obtaining the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment according to the two entry edge key points and the length of the target two-dimensional bounding box A; or, If the target two-dimensional bounding box A matches the three key points included in the corresponding parking space key point set, a rectangle formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or, If the target two-dimensional bounding box A matches the four key points included in the corresponding parking space key point set, a rectangle formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment; or, If the target two-dimensional bounding box A does not match any of the four key points included in its corresponding parking space key point set, the target two-dimensional bounding box A is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box A at the current moment.

11. The device according to claim 9, characterized in that In terms of performing oblique parking space post-processing according to the target two-dimensional bounding box B and the corresponding parking space key point set to obtain the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment, the obtaining unit is specifically used to: If the target two-dimensional bounding box B matches two entry edge key points included in the corresponding parking space key point set, then obtaining the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment according to the two entry edge key points, the parking space rotation angle corresponding to the target two-dimensional bounding box B and the length of the target two-dimensional bounding box B; or, If the target two-dimensional bounding box B matches the three key points included in the corresponding parking space key point set, a parallelogram formed by the three key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment; or, If the target two-dimensional bounding box B matches the four key points included in the corresponding parking space key point set, the parallelogram formed by the four key points is determined as the target parking space information of the target vehicle corresponding to the target two-dimensional bounding box B at the current moment.

12. The device according to claim 10 or 11, characterized in that The obtaining unit is further used to obtain the driving data of the target vehicle from the current moment to the next moment adjacent to the current moment; The obtaining unit is further used to update the position of the N target parking space information according to the driving data to obtain the N target parking space information after the position is updated; The obtaining unit is further used to obtain M target parking space information of the target vehicle at the next moment, where M is an integer greater than or equal to 1; The obtaining unit is further configured to set the parking space state corresponding to the target parking space information C to a newly added parking space if the target parking space information C does not match any of the N target parking space information after the position is updated, and the target parking space information C is any one of the M target parking space information; or, if the target parking space information D matches one of the N target parking space information after the position is updated, set the parking space state corresponding to the target parking space information D to an original parking space, and the target parking space information D is any one of the M target parking space information.

13. An electronic device, characterized in that: The electronic device comprises a processor, a communication interface and a memory coupled to each other, wherein: The processor is used to obtain a first fisheye stitched image of a target vehicle at a current moment, where the target vehicle is an autonomous driving vehicle traveling in a parking lot; The processor is further configured to input the first fisheye stitched image into a neural network model for processing, and output N groups of parking space data, N parking space key point sets, and N parking space types, wherein the N parking space key point sets and the N parking space types correspond to the N groups of parking space data one by one, respectively, and N is an integer greater than or equal to 1; The processor is further configured to obtain N initial parking space information of the target vehicle at the current moment according to the N groups of parking space data; The processor is further configured to correct the N initial parking space information according to the N parking space key point sets and the N parking space types to obtain the N target parking space information of the target vehicle at the current moment.

14. A computer-readable storage medium, characterized in that: A computer program is stored, and the computer program is executed by hardware to implement the method according to any one of claims 1 to 6.

15. A chip system, characterized in that: The chip system includes a processor for supporting the parking space detection device to implement the method described in any one of claims 1 to 6.

Images