Parking space detection method and device, terminal and computer readable storage medium
By acquiring peripheral vision images, circumferential vision images and lidar point cloud data in complex parking lot environments, combining panoramic circumferential vision system images, and using BEV model and dockable model for fusion analysis, the low scene coverage rate in the existing technology is solved. The problem of low detection accuracy achieves higher detection accuracy and coverage.
Patent Information
- Application Number
- CN202411019813.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art problems of low scene coverage and low detection accuracy of parking spaces in complex parking lot environments.
By acquiring peripheral vision images, circumferential vision images and lidar point cloud data, fusion is performed based on the BEV model, combined with the panoramic circumferential vision system image analysis, the parking space is analyzed using the parking space, and the parking space is output.
It improves the scene coverage rate and parking space detection accuracy in complex parking lot environments.
Smart Images

Figure CN119007435A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method, device, terminal and computer-readable storage medium for detecting available parking spaces. Background Art
[0002] In the field of unmanned driving and robotics, environmental perception and automatic parking in complex parking environments are among the most frequently used functions by users.
[0003] In the existing parking lot perception technology, the available parking spaces around the vehicle at the current moment can be output by inputting the surrounding image, surround image, LiDAR point cloud data, and panoramic surround system data. Since the available parking attributes in this parking lot perception technology need to be post-processed in combination with obstacles before the corresponding available parking spaces can be judged, and this part requires a lot of manual codes and manually designed rules, and there are many extreme cases, resulting in low scene coverage and low detection accuracy of available parking spaces in complex parking lot environments.
[0004] Therefore, the prior art needs to be improved. Summary of the invention
[0005] The technical problem to be solved by the present invention is that, in view of the defects of the prior art, the present invention provides a method, device, terminal and computer-readable storage medium for detecting available parking spaces, so as to solve the problems of low scene coverage and low accuracy of detecting available parking spaces in the prior art in a complex parking lot environment.
[0006] The technical solution adopted by the present invention to solve the technical problem is as follows:
[0007] In a first aspect, the present invention provides a method for detecting a parking space that can be parked, comprising:
[0008] Obtain the current panoramic image, surround image and lidar point cloud data;
[0009] Based on the BEV model, the surrounding image, the surround image and the laser radar point cloud data are fused to obtain a drivable area and a detection target in the parking lot;
[0010] Acquire a panoramic view system image, and analyze the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point;
[0011] According to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, the parkingable attributes of the parking spaces in the drivable area are analyzed using a parkingable model, and the parkingable attributes corresponding to each parking space are output.
[0012] In one implementation, the fusion of the surrounding image, the surround image and the lidar point cloud data based on the BEV model to obtain the drivable area and the detection target in the parking lot includes:
[0013] Performing feature conversion on the surrounding view image and the all-around view image to obtain a first BEV image feature;
[0014] Performing point cloud conversion on the laser radar point cloud data to obtain a second BEV image feature;
[0015] Based on the first BEV image feature and the second BEV image feature, a fusion is performed using a cross-attention mechanism to obtain the drivable area in the parking lot and the detection target.
[0016] In one implementation, the performing feature conversion on the surrounding view image and the all-around view image to obtain the first BEV image feature includes:
[0017] Inputting the surrounding view image and the all-around view image into a first image backbone network in the BEV model, and outputting 2D image features;
[0018] A BEV generation network based on the BEV model transforms the 2D image features into the first BEV image features.
[0019] In one implementation, performing point cloud conversion on the laser radar point cloud data to obtain a second BEV image feature includes:
[0020] Inputting the laser radar point cloud data into the laser radar backbone network of the BEV model to obtain a grouping feature map;
[0021] The sparse 3D point cloud features are converted into dense 3D BEV image features using a dispersion algorithm to obtain the second BEV image features.
[0022] In one implementation, the acquiring of a panoramic view system image and analyzing the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point include:
[0023] Acquire the panoramic view system image;
[0024] The panoramic surround view system image is input into the second image backbone network in the parking space model for analysis, and the parking space corner point and the parking space center point are output.
[0025] In one implementation, the parking space in the drivable area is analyzed using a parking model according to the drivable area in the parking lot, the detection target, the parking space corner point and the parking space center point, and the parking space corresponding to each parking space is output, including:
[0026] According to the detected parking space center point, a feature map of the same size is intercepted in the corresponding drivable area of the parking lot;
[0027] The intercepted feature map is input into the third image backbone network in the parking availability model to perform parking availability attribute analysis, and the parking availability attribute information of the corresponding parking space is output.
[0028] In one implementation, the method further includes:
[0029] The parking availability attribute information is used as the final output, and the drivable area in the parking lot, the detection target and the parking space information are used as the intermediate outputs, and end-to-end training is performed to obtain a network composed of the BEV model, the parking space model and the parking availability model connected in series.
[0030] In a second aspect, the present invention provides a parking space detection device, comprising:
[0031] The acquisition module is used to obtain the current panoramic image, surround image and lidar point cloud data;
[0032] A fusion module, used for fusing the surrounding image, the surround image and the laser radar point cloud data based on the BEV model to obtain a drivable area and a detection target in the parking lot;
[0033] A panoramic view system image analysis module, used to obtain a panoramic view system image, and analyze the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point;
[0034] The parking attribute analysis module is used to analyze the parking attributes of the parking spaces in the drivable area using the parking model according to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, and output the parking attributes corresponding to each parking space.
[0035] In a third aspect, the present invention provides a terminal, comprising: a processor and a memory, wherein the memory stores a parking space detection program, and when the parking space detection program is executed by the processor, it is used to implement the operation of the parking space detection method as described in the first aspect.
[0036] In a fourth aspect, the present invention further provides a medium, which is a computer-readable storage medium, and which stores a parking space detection program. When the parking space detection program is executed by a processor, it is used to implement the operation of the parking space detection method described in the first aspect.
[0037] The present invention adopts the above technical solution to achieve the following effects:
[0038] The present invention can obtain the surrounding image, surround image and laser radar point cloud data at the current moment, fuse the surrounding image, surround image and laser radar point cloud data based on the BEV model, and obtain the drivable area and detection target in the parking lot; and obtain the panoramic surround system image, and analyze the panoramic surround system image based on the parking space model to obtain the parking space corner point and parking space center point. According to the drivable area, detection target, parking space corner point and parking space center point in the parking lot, the parking space in the drivable area can be analyzed by using the parking model, and the parking properties corresponding to each parking space can be output. The present invention detects the drivable parking spaces by model fusion, which improves the scene coverage and the detection accuracy of the drivable parking spaces in a complex parking lot environment. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.
[0040] Figure 1 It is a flow chart of a method for detecting available parking spaces in one implementation of the present invention.
[0041] Figure 2 It is a structural diagram of a parking space detection model in one implementation of the present invention.
[0042] Figure 3 It is a processing flow chart of a parking space detection model in one implementation of the present invention.
[0043] Figure 4 It is a functional principle diagram of a terminal in one implementation of the present invention.
[0044] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION
[0045] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0046] Exemplary Methods
[0047] In the existing parking lot perception technology, the available parking spaces around the vehicle at the current moment can be output by inputting the surrounding image, surround image, LiDAR point cloud data, and panoramic surround system data. Since the available parking attributes in this parking lot perception technology need to be post-processed in combination with obstacles before the corresponding available parking spaces can be judged, and this part requires a lot of manual codes and manually designed rules, and there are many extreme cases, resulting in low scene coverage and low detection accuracy of available parking spaces in complex parking lot environments.
[0048] In view of the above technical problems, a method for detecting a parking space that can be parked is provided in an embodiment of the present invention. The method can fuse the surrounding image, surround image and lidar point cloud data at the current moment based on the BEV model to obtain the drivable area and detection target in the parking lot; and obtain the panoramic surround system image and analyze the panoramic surround system image based on the parking space model to obtain the parking space corner point and parking space center point. According to the drivable area, detection target, parking space corner point and parking space center point in the parking lot, the parking space model can be used to analyze the parking space in the drivable area, and the parking space corresponding to each parking space can be output. In the embodiment of the present invention, the parking space that can be parked is detected by model fusion, which improves the scene coverage rate and the detection accuracy of the parking space that can be parked in a complex parking lot environment.
[0049] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting a parking space that can be parked, comprising the following steps:
[0050] Step S100, obtaining the surrounding image, surround image and lidar point cloud data at the current moment.
[0051] In this embodiment, in a complex parking lot environment, due to factors such as the parking lot terrain, obstacles in the site, sensor sensing range, artificial codes, and numerous artificially designed rules, as well as many extreme cases, the scene coverage is low and the detection accuracy of available parking spaces is low in a complex parking lot environment; therefore, in this embodiment, the BEV model, parking space model, and available parking model are fused by model fusion, so that the available parking spaces are detected by the fused model, which can improve the scene coverage and the detection accuracy of available parking spaces in a complex parking lot environment.
[0052] Specifically, the present embodiment provides a solution for realizing end-to-end parking space detection in a parking lot based on laser radar, images and panoramic surround view system images (i.e., AVM images). The solution provided by the embodiment of the present invention only needs to input the surrounding view images, surround view images, laser radar point cloud data and panoramic surround view system images at the current moment, and can output the precise location of the parking space in the parking lot and the attribute information of whether the corresponding parking space can be parked, thereby realizing the automatic parking function in a complex parking lot environment based on the output information.
[0053] In this embodiment, before detecting the available parking spaces, it is necessary to obtain the surrounding view images, surround view images and lidar point cloud data at the current moment; as an example, the surrounding view images in this embodiment are collected by multiple surround view cameras installed in a circle around the vehicle, and the surround view cameras are wide-angle cameras with a shooting angle of about 100°; the surround view images are collected by surround view fisheye cameras installed under the left and right rearview mirrors and under the front and rear license plates of the vehicle, and the surround view fisheye cameras have a large viewing angle of more than 180° and better perception of close distances; the lidar point cloud data are collected by a lidar installed on the top of the vehicle.
[0054] In this embodiment, the surrounding view image, surround view image and lidar point cloud data acquired at the current moment are analyzed by using a parking space detection model to obtain information on whether the parking spaces around the vehicle at the current moment are available for parking; wherein the parking space detection model is a fusion model obtained by model fusion.
[0055] like Figure 2 As shown, the parking space detection model includes: a BEV model, a parking space model and a parking model; the BEV model, the parking space model and the parking model are combined into a whole.
[0056] Specifically, Figure 2 and Figure 3As shown, the input of the BEV model is the image data collected by the camera (i.e., the surrounding image and the surround image at the current moment) and the data collected by the laser radar (i.e., the laser radar point cloud data); the output is the output result of the fusion module, i.e., the drivable area in the parking lot and the detection target output by the fusion module (the output is the obstacle information around the vehicle at the current moment); the input of the parking space model is the data of the panoramic surround view system (i.e., the panoramic surround view system image), and the output is the parking space and its corresponding parking space corner point, parking space center point and other information, that is, the output is the information of all parking spaces around the vehicle at the current moment; the input of the parking model is the output of the BEV model (i.e., the drivable area in the parking lot, the detection target) and the output of the parking space model (i.e., the parking space corner point and parking space center point and other information), and the output is the attribute information of whether the parking spaces around the vehicle at the current moment are parkable.
[0057] like Figure 1 As shown, in one implementation of the embodiment of the present invention, the method for detecting available parking spaces further includes the following steps:
[0058] Step S200: Based on the BEV model, the surrounding view image, the surround view image and the lidar point cloud data are fused to obtain a drivable area and a detection target in the parking lot.
[0059] like Figure 2 As shown, in this embodiment, the BEV model includes: a first image backbone network, a lidar backbone network, a BEV generation network (bevformer), a 3D point cloud conversion module and a fusion module; wherein the lidar backbone network adopts a point++ structure as the backbone network.
[0060] like Figure 2 and Figure 3 As shown, in this embodiment, the input of the first image backbone network is the surrounding image and the all-around image collected by the camera at the current moment, and the output is 2D image features; the 2D image features are converted from 2D image features to 3D image features through the BEV generation network (bevformer), thereby obtaining the features after 2D conversion to 3D; the input of the lidar backbone network is the lidar point cloud data collected by the lidar, and the output is a grouped feature map; the grouped feature map is converted through the 3D point cloud conversion module, and the output is the features after sparse conversion to dense.
[0061] The features after 2D conversion to 3D output by the BEV generation network and the features after sparse conversion to dense output by the 3D point cloud conversion module are input into a fusion module, which then outputs perception information such as the drivable area (freespace) and the detected target (detection), thereby obtaining the obstacle information around the vehicle at the current moment.
[0062] Specifically, in an implementation of this embodiment, step S200 includes the following steps:
[0063] Step S201 , performing feature conversion on the surrounding view image and the all-around view image to obtain a first BEV image feature.
[0064] In one implementation of this embodiment, step S201 includes the following steps:
[0065] Step S201a, inputting the surrounding view image and the all-around view image into the first image backbone network in the BEV model, and outputting 2D image features;
[0066] Step S201b: converting the 2D image features into the first BEV image features based on the BEV generation network of the BEV model.
[0067] like Figure 3 As shown, in this embodiment, after the surrounding view image and the all-around view image at the current moment are input, these images are passed through the first image backbone network in the BEV model to extract image features, and 2D image features are output; for these 2D image features, they are input into the BEV generation network for feature conversion, so as to obtain the features after 2D conversion to 3D; in the BEV generation network, the bevformer (i.e., bev perception) method is used to convert the 2D image features into BEV image features (i.e., first BEV image features).
[0068] Specifically, in an implementation of this embodiment, step S200 further includes the following steps:
[0069] Step S202, performing point cloud conversion on the laser radar point cloud data to obtain a second BEV image feature.
[0070] In one implementation of this embodiment, step S202 includes the following steps:
[0071] Step S202a, inputting the laser radar point cloud data into the laser radar backbone network of the BEV model to obtain a grouping feature map;
[0072] Step S202b: using a dispersion algorithm to convert the sparse 3D point cloud features into dense 3D BEV image features to obtain the second BEV image features.
[0073] like Figure 3As shown, in this embodiment, after the lidar point cloud at the current moment is input, these lidar point clouds pass through the lidar backbone network of the BEV model (i.e. point++ is used as the backbone network) to output a grouped feature map; for the grouped feature map, the 3D point cloud conversion module is input for conversion to obtain the features after sparse conversion to dense; in the conversion process of the 3D point cloud conversion module, a dispersed algorithm (i.e., a scatter algorithm) can be used to convert the sparse 3D point cloud features into dense 3D BEV image features.
[0074] Specifically, in an implementation of this embodiment, step S200 further includes the following steps:
[0075] Step S203: Based on the first BEV image feature and the second BEV image feature, a cross-attention mechanism is used to fuse them to obtain the drivable area in the parking lot and the detection target.
[0076] like Figure 3 As shown, in this embodiment, the features after 2D conversion to 3D output by the BEV generation network and the features after sparse conversion to dense output by the 3D point cloud conversion module are input into a fusion module for fusion to obtain the perception information of the drivable area and detection targets in the parking lot, that is, the obstacle information around the vehicle at the current moment is obtained.
[0077] As an example, in the fusion module, fusion is performed by means of a cross-attention mechanism, and the following perception results are output:
[0078] 1) Freespace: This area contains static obstacles (e.g., stationary cars, bushes, wheel chocks, ground locks, stationary bicycles, sports bicycles, walls, cones, curbs, and pillars). Static obstacles in the freespace can be accumulated and perceived in time series using a grid map algorithm (i.e., gridmap).
[0079] Specifically, the perception method of the grid map is: converting the characteristic map of the drivable area at the previous moment into the coordinate system of the current moment by means of pose (positioning information) changes, and fusing it through Kalman filtering to obtain the perception result of the drivable area; that is, obtaining the perception result of the drivable area and the static obstacles therein.
[0080] 2) Detection: Movable obstacles in the parking lot (e.g., cars, trucks, people, bicycles) are defined as detection targets; these movable obstacles can be tracked over a long time series using TrackFormer, an end-to-end multi-target tracking and segmentation model based on an encoder-decoder converter structure.
[0081] In this embodiment, by inputting the surrounding view image, surround view image, and lidar data into the BEV model, the BEV model can be used to output obstacle information around the vehicle at the current moment, so as to facilitate the subsequent analysis of whether the parking space can be parked through the parking availability model.
[0082] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting a parking space that can be parked, comprising the following steps:
[0083] Step S300: Acquire a panoramic view system image, and analyze the panoramic view system image based on a parking space model to obtain parking space corner points and a parking space center point.
[0084] In this embodiment, when the surrounding view image, the surround view image and the laser radar point cloud data are fused based on the BEV model, the panoramic surround view system image can also be analyzed based on the parking space model to obtain the parking space position and the corresponding parking space corner point and parking space center point and other information (that is, to obtain the information of all parking spaces around the vehicle at the current moment), so as to further analyze whether the current parking space can be parked according to the parking space information and the perception information output by the BEV model; Figure 2 As shown, the parking space model includes: a second image backbone network; Figure 3 As shown, the input of the second image backbone network is the panoramic view system image, and the output is the parking space position and the corresponding parking space corner points and parking space center points and other information.
[0085] Specifically, in an implementation of this embodiment, step S300 includes the following steps:
[0086] Step S301, acquiring the panoramic surround view system image;
[0087] Step S302: input the panoramic view system image into the second image backbone network in the parking space model for analysis, and output the parking space corner point and the parking space center point.
[0088] In this embodiment, the panoramic view system is usually composed of four cameras installed in the front, rear, left and right directions of the vehicle, and an additional bird's-eye view camera is included when necessary. The images captured by these cameras are combined into a panoramic view; by acquiring the panoramic view system image, the panoramic view system image is input into the second image backbone network in the parking space model for analysis, the parking space in the panoramic view system image, and the coordinate information of the parking space corner point and parking space center point corresponding to each parking space are determined, and the information of all parking spaces around the vehicle at the current moment is obtained; wherein the determined parking space in the panoramic view system image can be a line parking space or a space parking space.
[0089] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting a parking space that can be parked, comprising the following steps:
[0090] Step S400 , according to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, a parking availability model is used to analyze the parking availability attributes of the parking spaces in the drivable area, and the parking availability attributes corresponding to each parking space are output.
[0091] In this embodiment, the parking model is used to analyze the parking properties of the parking spaces in the drivable area, such as Figure 3 As shown, the input of the parking model is the output of the BEV model and the parking space model, and the output of the parking model is the parking attribute information corresponding to each parking space; Figure 2 As shown, in one embodiment, the dockable model includes: a third image backbone network.
[0092] Specifically, in an implementation of this embodiment, step S400 includes the following steps:
[0093] Step S401, based on the detected parking space center point, extract a feature map of the same size in the corresponding drivable area of the parking lot;
[0094] Step S402: Input the intercepted feature image into the third image backbone network in the parking availability model to perform parking availability attribute analysis, and output parking availability attribute information of the corresponding parking space.
[0095] In this embodiment, the drivable area in the parking lot, the detection target output by the BEV model, and the parking space corner point, parking space center point and other information output by the parking space model are used as inputs of the third image backbone network, and the parking attribute analysis is performed through the third image backbone network to output the parking attribute information of the corresponding parking space.
[0096] In the process of analyzing whether a parking space is available for parking by the third image backbone network, the third image backbone network uses the midpoint position of the parking space as a query for obstacle information around the vehicle at the current moment output by the BEV model and information about all parking spaces around the vehicle at the current moment output by the parking space model, crops a feature map of a fixed size from the output of the BEV model, and then inputs the new feature map obtained above into the parking available model to determine whether the parking space is available for parking, thereby outputting attribute information corresponding to the parking spaces around the vehicle at the current moment.
[0097] An embodiment of the present invention provides a method for detecting a parking space that can be parked, comprising the following steps:
[0098] Step S500, taking the parking availability attribute information as the final output, and taking the drivable area in the parking lot, the detection target and the parking space information as the intermediate output, performing end-to-end training, and obtaining a network composed of the BEV model, the parking space model and the parking availability model connected in series.
[0099] In this embodiment, during the construction and training of the available parking space detection model, the above models (i.e., the BEV model, the parking space model, and the available parking model) are connected in series to the network, and whether the vehicle can be parked is used as the final output, and information such as the drivable area (freespace) in the parking lot, the detection target (detection), and the parking space (parking space corner point, parking space center point) are used as the intermediate output of the available parking space detection model. The neural network is trained end-to-end to obtain a trained available parking space detection model; the trained available parking space detection model is used to detect the parking spaces around the vehicle in a complex parking lot environment.
[0100] As an example, in this embodiment, in addition to analyzing whether parking is possible through the parking space detection model, the corresponding attribute information can also be directly obtained through large model analysis; that is, from the raw data input from the sensor, through a large number of Transformer modules, the parking space can be directly output, and all intermediate results are removed.
[0101] Compared with existing automatic parking solutions in parking lots, the parking attributes in the existing solutions are all based on artificial rules, which cannot cover all scenarios and are limited by the obstacle detection accuracy. In this embodiment, data-driven problem solving is no longer solved by artificially constructed rules, thereby improving the detection accuracy of available parking spaces in complex perception environments.
[0102] This embodiment achieves the following technical effects through the above technical solution:
[0103] This embodiment acquires the surrounding image, surround image and LiDAR point cloud data at the current moment, and fuses the surrounding image, surround image and LiDAR point cloud data based on the BEV model to obtain the drivable area and detection target in the parking lot; and acquires the panoramic surround system image and analyzes the panoramic surround system image based on the parking space model to obtain the parking space corner point and parking space center point. According to the drivable area, detection target, parking space corner point and parking space center point in the parking lot, the parking space model can be used to analyze the parking properties of the parking spaces in the drivable area, and output the parking properties corresponding to each parking space. This embodiment detects the drivable parking spaces by model fusion, which improves the scene coverage and the detection accuracy of the drivable parking spaces in complex parking lot environments.
[0104] Exemplary Devices
[0105] Based on the above embodiments, the present invention further provides a parking space detection device, comprising:
[0106] The acquisition module is used to obtain the current panoramic image, surround image and lidar point cloud data;
[0107] A fusion module, used for fusing the surrounding image, the surround image and the laser radar point cloud data based on the BEV model to obtain a drivable area and a detection target in the parking lot;
[0108] A panoramic view system image analysis module, used to obtain a panoramic view system image, and analyze the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point;
[0109] The parking attribute analysis module is used to analyze the parking attributes of the parking spaces in the drivable area using the parking model according to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, and output the parking attributes corresponding to each parking space.
[0110] This embodiment achieves the following technical effects through the above technical solution:
[0111] This embodiment acquires the surrounding image, surround image and LiDAR point cloud data at the current moment, and fuses the surrounding image, surround image and LiDAR point cloud data based on the BEV model to obtain the drivable area and detection target in the parking lot; and acquires the panoramic surround system image and analyzes the panoramic surround system image based on the parking space model to obtain the parking space corner point and parking space center point. According to the drivable area, detection target, parking space corner point and parking space center point in the parking lot, the parking space model can be used to analyze the parking properties of the parking spaces in the drivable area, and output the parking properties corresponding to each parking space. This embodiment detects the drivable parking spaces by model fusion, which improves the scene coverage and the detection accuracy of the drivable parking spaces in complex parking lot environments.
[0112] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be as follows: Figure 4 shown.
[0113] The terminal includes: a processor, a memory, an interface, a display screen and a communication module connected through a system bus; wherein the processor of the terminal is used to provide computing and control capabilities; the memory of the terminal includes a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of the operating system and the computer program in the storage medium; the interface is used to connect to external devices; the display screen is used to display corresponding information; and the communication module is used to communicate with a cloud server or other devices.
[0114] When the computer program is executed by a processor, it is used to implement the operation of the parking space detection method.
[0115] It can be understood by those skilled in the art that Figure 4 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0116] In one embodiment, a terminal is provided, which includes: a processor and a memory, wherein the memory stores a parking space detection program, and when the parking space detection program is executed by the processor, it is used to implement the operation of the parking space detection method as described above.
[0117] In one embodiment, a storage medium is provided, wherein the storage medium stores a parking space detection program, and when the parking space detection program is executed by a processor, it is used to implement the operation of the parking space detection method as described above.
[0118] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and volatile memory.
[0119] In summary, the present invention provides a method, device, terminal and computer-readable storage medium for detecting available parking spaces, the method comprising: obtaining a surrounding image, surround image and laser radar point cloud data at the current moment; fusing the surrounding image, surround image and laser radar point cloud data based on the BEV model to obtain a drivable area and a detection target in the parking lot; obtaining a panoramic surround view system image, and analyzing the panoramic surround view system image based on a parking space model to obtain parking space corner points and parking space center points; according to the drivable area in the parking lot, the detection target, parking space corner points and parking space center points, using the parkable model to analyze the parking spaces in the drivable area for parkable attributes, and outputting the corresponding parking attributes of each parking space. The present invention detects available parking spaces by means of model fusion, thereby improving the scene coverage rate and the detection accuracy of available parking spaces in complex parking lot environments.
[0120] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.
Claims
1. A method for detecting a parking space available for parking, characterized in that: include: Obtain the current panoramic image, surround image and lidar point cloud data; Based on the BEV model, the surrounding image, the surround image and the laser radar point cloud data are fused to obtain a drivable area and a detection target in the parking lot; Acquire a panoramic view system image, and analyze the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point; According to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, the parkingable attributes of the parking spaces in the drivable area are analyzed using a parkingable model, and the parkingable attributes corresponding to each parking space are output.
2. The method for detecting available parking spaces according to claim 1, characterized in that: The method of fusing the surrounding image, the surround image and the laser radar point cloud data based on the BEV model to obtain a drivable area and a detection target in the parking lot includes: Performing feature conversion on the surrounding view image and the all-around view image to obtain a first BEV image feature; Performing point cloud conversion on the laser radar point cloud data to obtain a second BEV image feature; Based on the first BEV image feature and the second BEV image feature, a fusion is performed using a cross-attention mechanism to obtain the drivable area in the parking lot and the detection target.
3. The method for detecting available parking spaces according to claim 2, characterized in that: The performing feature conversion on the surrounding view image and the all-around view image to obtain a first BEV image feature includes: Inputting the surrounding view image and the all-around view image into a first image backbone network in the BEV model, and outputting 2D image features; A BEV generation network based on the BEV model transforms the 2D image features into the first BEV image features.
4. The method for detecting available parking spaces according to claim 2, characterized in that: The step of performing point cloud conversion on the laser radar point cloud data to obtain a second BEV image feature includes: Inputting the laser radar point cloud data into the laser radar backbone network of the BEV model to obtain a grouping feature map; The sparse 3D point cloud features are converted into dense 3D BEV image features using a dispersion algorithm to obtain the second BEV image features.
5. The method for detecting available parking spaces according to claim 1, characterized in that: The step of acquiring a panoramic view system image and analyzing the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point includes: Acquire the panoramic view system image; The panoramic surround view system image is input into the second image backbone network in the parking space model for analysis, and the parking space corner point and the parking space center point are output.
6. The method for detecting available parking spaces according to claim 1, characterized in that: The method of analyzing the parking properties of parking spaces in the drivable area using a parking model according to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, and outputting the parking properties corresponding to each parking space includes: According to the detected parking space center point, a feature map of the same size is intercepted in the corresponding drivable area of the parking lot; The intercepted feature map is input into the third image backbone network in the parking availability model to perform parking availability attribute analysis, and the parking availability attribute information of the corresponding parking space is output.
7. The method for detecting available parking spaces according to claim 1, characterized in that: The method further comprises: The parking availability attribute information is used as the final output, and the drivable area in the parking lot, the detection target and the parking space information are used as the intermediate outputs, and end-to-end training is performed to obtain a network composed of the BEV model, the parking space model and the parking availability model connected in series.
8. A parking space detection device, characterized in that: include: The acquisition module is used to obtain the current panoramic image, surround image and lidar point cloud data; A fusion module, used for fusing the surrounding image, the surround image and the laser radar point cloud data based on the BEV model to obtain a drivable area and a detection target in the parking lot; A panoramic view system image analysis module, used to obtain a panoramic view system image, and analyze the panoramic view system image based on a parking space model to obtain a parking space corner point and a parking space center point; The parking attribute analysis module is used to analyze the parking attributes of the parking spaces in the drivable area using the parking model according to the drivable area in the parking lot, the detection target, the parking space corner points and the parking space center points, and output the parking attributes corresponding to each parking space.
9. A terminal, characterized in that: include: A processor and a memory, wherein the memory stores a parking space detection program, and when the parking space detection program is executed by the processor, it is used to implement the operation of the parking space detection method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a parking space detection program, and when the parking space detection program is executed by the processor, it is used to implement the operation of the parking space detection method according to any one of claims 1 to 7.
Citation Information
Cited By
Display control method and device and electronic equipment
CN120621223A
Vehicle parking method, electronic equipment and storage medium
CN121106198A
