Parking control method and device and vehicle

Abstract

The application provides a parking control method and device and a vehicle. The method comprises the following steps: acquiring a panoramic image in a driving range of a vehicle to be parked; processing the panoramic image to obtain effective parking space information, parking line information and a drivable area; determining target parking space coordinate information according to the effective parking space information and the parking line information; determining a target parking path according to the drivable area; and controlling the vehicle to be parked according to the target parking space coordinate information and the target parking path. The application can avoid parking position deviation, make the parking path more reasonable, reduce the number of times of parking garage, improve the parking efficiency and improve the user parking experience.

Description

Technical Field

[0001] This invention relates to the field of vehicle technology, and in particular to a parking control method, device, and vehicle. Background Technology

[0002] In related technologies, during automatic parking / assisted parking, parking position deviation and multiple instances of merging into the parking space due to unreasonable parking path planning are prone to occur, resulting in low parking efficiency and reduced user parking experience. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0004] Therefore, one objective of this invention is to propose a parking control method that avoids parking position deviation, makes the parking path more reasonable, reduces the number of times the car needs to be maneuvered, improves parking efficiency, and enhances the user's parking experience.

[0005] Therefore, a second objective of the present invention is to provide a parking control device.

[0006] Therefore, a third objective of the present invention is to provide a vehicle.

[0007] To achieve the above objectives, a first aspect of the present invention provides a parking control method, the method comprising: acquiring a panoramic image of the driving range of a vehicle to be parked; processing the panoramic image to obtain effective parking space information, parking line information, and a drivable area; determining target parking space coordinate information based on the effective parking space information and the parking line information; determining a target parking path based on the drivable area; and controlling the vehicle to be parked based on the target parking space coordinate information and the target parking path.

[0008] According to the parking control method of the present invention, when a vehicle is parking, a panoramic image is processed to obtain effective parking space information, parking line information, and drivable area. The parking line information and drivable area are taken into account to optimize the parking space coordinate information and parking path, thereby obtaining the target parking space coordinate information and target parking path. The vehicle is controlled to park based on the target parking space coordinate information and target parking path, avoiding parking position deviation and making the parking path more reasonable, thereby reducing the number of times the vehicle needs to maneuver in the parking space. This improves parking efficiency and enhances the user's parking experience.

[0009] In some embodiments, processing the panoramic image to obtain effective parking space information includes: extracting feature vectors from the panoramic image, determining multidimensional coordinate information of parking spaces and two-dimensional coordinate information of obstacles in the panoramic image, wherein the multidimensional coordinate information includes the rotation angle of the parking space; and obtaining the effective parking space information based on the multidimensional coordinate information and the two-dimensional coordinate information.

[0010] In some embodiments, outputting the valid parking space information based on the multidimensional coordinate information and the two-dimensional coordinate information includes: determining the parking space boundary frame of the parking space based on the multidimensional coordinate information; determining the obstacle boundary frame of the obstacle and the center point coordinates of the obstacle based on the two-dimensional coordinate information; determining whether the center point coordinates of the obstacle are within the parking space boundary frame of the parking space; if not, determining the boundary overlap ratio of the parking space boundary frame and the obstacle boundary frame; when the boundary overlap ratio does not exceed a preset ratio threshold, determining the parking space as a valid parking space and outputting the valid parking space information.

[0011] In some embodiments, processing the panoramic image to obtain the parking line information includes: extracting feature vectors from the panoramic image, determining a set of parking line pixels in the panoramic image, and obtaining the parking line information based on the set of parking line pixels.

[0012] In some embodiments, processing the panoramic image to obtain the drivable area includes: extracting feature vectors from the panoramic image and determining the pixels of the drivable area in the panoramic image; and determining the drivable area based on the pixels of the drivable area.

[0013] In some embodiments, determining the target parking space coordinate information based on the valid parking space information and the parking space line information includes: correcting the area and angle of the parking space boundary box of the valid parking space based on the parking space line information to obtain the target parking space coordinate information.

[0014] In some embodiments, determining a target parking path based on the drivable area includes: determining multiple parking paths in the drivable area; and determining the target parking path based on the multiple parking paths.

[0015] In some embodiments, determining multiple parking paths in the drivable area includes: obtaining the current location information of the parked vehicle; and determining multiple parking paths based on the current location information and the target parking space coordinate information.

[0016] In some embodiments, determining the target parking path based on multiple parking paths includes: obtaining a set of probability values ​​for successful parking on multiple parking paths, and taking the path with the highest probability value as the target parking path.

[0017] To achieve the above objectives, a second aspect of the present invention provides a parking control device, the device comprising: a processor, a memory, and a parking control program stored in the memory and executable on the processor, wherein the parking control program, when executed by the processor, implements the parking control method as described in the above embodiments.

[0018] According to an embodiment of the present invention, when a vehicle is parking, the parking control device processes a panoramic image to obtain effective parking space information, parking line information, and drivable area. Taking the parking line information and drivable area into account, the device optimizes the parking space coordinate information and parking path to obtain target parking space coordinate information and target parking path. Based on the target parking space coordinate information and target parking path, the device controls the vehicle's parking to avoid parking position deviation, making the parking path more reasonable, reducing the number of times the vehicle needs to maneuver in the parking space, and improving both parking efficiency and user parking experience.

[0019] To achieve the above objectives, a third aspect of the present invention provides a vehicle comprising: a parking control device as described in the above embodiments.

[0020] According to an embodiment of the present invention, when the vehicle is parked, the panoramic image is processed to obtain effective parking space information, parking line information, and drivable area. The parking line information and drivable area are taken into account to optimize the parking space coordinate information and parking path, thereby obtaining the target parking space coordinate information and target parking path. This avoids parking position deviation, makes the parking path more reasonable, reduces the number of times the vehicle needs to maneuver in the parking space, increases parking efficiency, and improves the user's parking experience.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0023] Figure 1 This is a flowchart of an autonomous parking assistance method based on panoramic bird's-eye view and deep learning in related technologies;

[0024] Figure 2 This is a flowchart of a parking method based on parking space calibration in related technologies;

[0025] Figure 3 This is a flowchart of a parking control method according to an embodiment of the present invention;

[0026] Figure 4 This is a flowchart of acquiring a panoramic image within the driving range of a vehicle to be parked, according to an embodiment of the present invention;

[0027] Figure 5(a) is a schematic diagram of the unimproved target detection module algorithm;

[0028] Figure 5(b) is a schematic diagram of the detection algorithm of the improved target detection module;

[0029] Figure 6 This is a schematic diagram of a parking control method according to an embodiment of the present invention;

[0030] Figure 7 This is a flowchart of a parking control method according to an embodiment of the present invention;

[0031] Figure 8 This is a block diagram of a vehicle according to an embodiment of the present invention. Detailed Implementation

[0032] The embodiments of the present invention are described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. The embodiments of the present invention are described in detail below.

[0033] In related technologies, such as Figure 1 The diagram shows a flowchart of a current autonomous parking assistance method based on panoramic bird's-eye view and deep learning. The vehicle uses object detection and semantic segmentation algorithms to detect parking spaces, drivable areas, obstacles, and pedestrians, obtaining map and object location information to plan the parking path. However, using dual deep learning models to detect this information places high demands on onboard computing power, and detecting a single image twice is slow, increasing parking costs and reducing user experience.

[0034] like Figure 2 The diagram shows a flowchart of a current parking method based on parking space calibration. The vehicle uses a preset splicing strategy to calibrate the parking space error, but it does not detect the parking lines, resulting in unreasonable parking path planning. This leads to situations where the vehicle crosses the lines, the parking position deviates, and multiple instances of "skimming" the parking space, resulting in an unsatisfactory parking outcome.

[0035] Therefore, the parking control method of this invention improves the algorithm based on a single model of multi-task deep learning, and detects effective parking space information, parking lines and drivable areas. When determining the target parking space coordinates and the target parking path, the parking line information and drivable area are taken into account, making the parking path more reasonable, reducing the number of times parking is traversed, and improving parking efficiency.

[0036] The following is combined Figures 3-7 The parking control method of this invention will be illustrated by example.

[0037] like Figure 3 As shown, the parking control method of this embodiment includes at least steps S1-S5.

[0038] Step S1: Obtain a panoramic image of the driving range of the vehicle to be parked.

[0039] Among them, the panoramic image is a panoramic image obtained by the vehicle's panoramic imaging module after distortion correction, viewpoint transformation, fusion and stitching of the images collected by the four surround view cameras. By obtaining panoramic images within the driving range of the vehicle to be parked, it can be used for analysis by a multi-task deep learning model.

[0040] In an embodiment, such as Figure 4 The diagram shows a flowchart of acquiring a panoramic image within the driving range of a vehicle to be parked, according to an embodiment of the present invention. An image acquisition module, such as a fisheye camera mounted on the front and rear bumpers and left and right rearview mirrors, acquires images of the surrounding environment during the vehicle's movement, resulting in four fisheye images. After obtaining the fisheye images, image distortion correction is performed. First, each pixel in the four fisheye images is processed, and the corresponding normalized imaging plane mapping point is calculated using intrinsic parameter information. Then, the corresponding mapping point is calculated using the fisheye camera's intrinsic parameters and distortion coefficients. Finally, an interpolation algorithm is used to obtain the pixel values ​​of the mapped image, resulting in the corrected image. A projective transformation is then performed on the corrected image to convert it into a bird's-eye view.

[0041] After obtaining four bird's-eye view images, the overlapping parts of the images are stitched together, and a weighted smoothing algorithm is used to handle problems such as stitching seams and ghosting to obtain a 360° panoramic image for information extraction by a multi-task deep learning model.

[0042] For example, such as Figure 4 The diagram shows a flowchart of acquiring a panoramic image within the driving range of a vehicle to be parked, according to an embodiment of the present invention. The steps for acquiring the panoramic image include at least steps S11-S15.

[0043] Step S11: Collect images of the surrounding environment during vehicle movement to obtain four fisheye images.

[0044] Step S12: Perform distortion correction processing on the four fisheye images to obtain the corrected images.

[0045] Step S13: Perform a projective transformation on the corrected image to obtain four bird's-eye view images.

[0046] Step S14: The four bird's-eye view images are merged and stitched together to obtain a 360° panoramic image.

[0047] By acquiring panoramic images, it is easier to process them for subsequent parking control.

[0048] Step S2: Process the panoramic image to obtain valid parking space information, parking line information, and drivable area.

[0049] The processing of panoramic images involves using a single model to handle multiple tasks through multi-task deep learning, enriching the detection information. For example, the multi-task deep learning model includes a target detection module, a parking line detection module, and a drivable area detection module. The target detection module is responsible for detecting parking spaces and obstacles, the parking line detection module is responsible for detecting parking lines, and the drivable area module is responsible for detecting the drivable area of ​​the vehicle during parking. The multi-task deep learning model is pre-trained using panoramic images, and can be used for detection simply by inputting a panoramic image. It has low computational requirements, accurate and simple information fusion, and can improve detection speed and reduce detection costs.

[0050] In this embodiment, after acquiring a panoramic image of the driving range of the vehicle to be parked, the multi-task deep learning module performs multi-task processing on the input panoramic image, such as information extraction, and sends the extracted information to the target detection module to obtain valid parking space information, such as parking space and obstacle information; at the same time, it sends it to the parking line detection module to obtain parking line information; and at the same time, it sends it to the drivable area detection module to obtain the drivable area, such as the area of ​​the drivable area, so as to determine the target parking space coordinate information and the target parking path based on the valid parking space information, parking line information and drivable area.

[0051] Step S3: Determine the target parking space coordinates based on the valid parking space information and parking line information.

[0052] Among them, the target parking space coordinate information is the location information of the parking space where the vehicle needs to park. By using the valid parking space information and parking line information, more accurate target parking space coordinate information can be determined.

[0053] In this embodiment, the multi-task deep learning model obtains valid parking space information through the target detection module and parking space line information through the parking space line detection module. Then, it fuses the valid parking space information and the parking space line information and maps this information onto the panoramic image. Based on the valid parking space information and the parking space line information, it continuously corrects the size and angle information of the parking space bounding box in the panoramic image to obtain more accurate valid parking space coordinate information.

[0054] Understandably, detecting parking space lines can prevent vehicles from crossing the lines or tilting after parking, ensuring that the vehicle is centered on the parking space line and that the information is integrated with valid parking space information to obtain more accurate target parking space coordinates.

[0055] Step S4: Determine the target parking route based on the drivable area.

[0056] The target parking path is the optimal parking path for a vehicle to enter the target parking space. By determining the path with a high success rate within the drivable area, the success rate of vehicle parking can be improved.

[0057] In this embodiment, after obtaining the drivable area through the drivable area detection module, the multi-task deep learning model outputs drivable area information for parking route planning. Within the drivable area, multiple parking paths to the target parking space are determined. After determining multiple parking paths to the target parking space, the parking path with the higher success rate is selected as the target parking path to improve the success rate of vehicle parking.

[0058] Understandably, when the drivable area detection module detects drivable areas on the road, it can combine information such as vehicles, obstacles, pedestrians, and parked vehicles to plan parking routes, avoiding multiple instances of vehicles circling into parking spaces during automatic or assisted parking, and improving the efficiency of vehicle parking.

[0059] Step S5: Control the parking of the vehicle to be parked based on the target parking space coordinates and the target parking path.

[0060] In this embodiment, after the multi-task deep learning model determines the target parking space coordinates and the target parking path, it sends the target parking space coordinates and the target parking path to the vehicle. The vehicle maps the target parking space coordinates and the target parking path onto the panoramic image. When the vehicle is parking, it adjusts the vehicle's status and the parking space based on the target parking space coordinates and the target parking path to optimize the parking route, thereby reducing the number of times the vehicle has to maneuver in the parking space and the parking time, and improving the success rate of parking.

[0061] According to the parking control method of this invention, when a vehicle is parking, the panoramic image is processed to obtain effective parking space information, parking line information, and drivable area. The parking line information and drivable area are taken into account to optimize the parking space coordinate information and parking path, resulting in target parking space coordinate information and target parking path. This avoids parking position deviations, makes the parking path more reasonable, reduces the number of times parking spaces need to be traversed, and improves parking efficiency. Furthermore, when determining the above information, a multi-task deep learning model is used to process the panoramic image. Compared to a combination of two deep learning models, this method has lower computational requirements, more accurate and simpler feature fusion, and eliminates the need for double detection of a single image, increasing detection speed. This reduces parking costs while improving user experience.

[0062] Currently, when parking vehicles detect parking spaces, the algorithm itself has limitations. It performs well in detecting parallel and perpendicular parking spaces, but poorly in detecting angled parking spaces. This is mainly because the current detection of parking spaces can only be performed in two dimensions: width and height, and cannot be rotated. Therefore, the target detection module in this embodiment of the invention adds rotation angle dimension information when determining valid parking space information to improve the detection accuracy of angled parking spaces.

[0063] In some embodiments, when processing the panoramic image to obtain valid parking space information, feature vector extraction is performed on the panoramic image to determine the multidimensional coordinate information of the parking space and the two-dimensional coordinate information of the obstacle in the panoramic image. The multidimensional coordinate information includes the rotation angle of the parking space. Valid parking space information is obtained based on the multidimensional coordinate information and the two-dimensional coordinate information.

[0064] Understandably, when processing panoramic images, the panoramic images are input into the target detection module. In order to improve the detection accuracy of angled parking spaces, the target detection module improves the algorithm. When determining valid parking space information, in addition to detecting the width and height of the parking space, it adds the detection of the rotation angle of the parking space, thereby improving the vehicle's ability to identify different parking spaces.

[0065] In this embodiment, after acquiring a panoramic image of the driving range of the vehicle to be parked, the convolutional neural network module uses a convolutional algorithm to extract feature vectors from the panoramic image to obtain feature vectors. The extracted feature vectors are then sent to the target detection module so that the target detection module can detect parking spaces and obstacles, such as pedestrians, vehicles, limit bars, ground locks in open / closed states, wheel chocks, and traffic cones.

[0066] After receiving the feature vector, the target detection module classifies and regresses the feature vector and determines the information category to which the feature vector belongs. For example, it determines that the feature vector belongs to the obstacle category and determines the two-dimensional coordinate information of the obstacle in the panoramic image. For example, it calculates the two-dimensional coordinate information (x, y, w, h) of the obstacle in the panoramic image through a regression equation, where (x, y) are the coordinates of the center point of the obstacle in the panoramic image, and (w, h) are the width and height of the obstacle.

[0067] If the feature vector is determined to belong to the parking space category, then the multi-dimensional coordinate information of the parking space in the panoramic image is determined simultaneously. For example, the regression branch of the object detection algorithm is improved by adding the parking space angle dimension information, that is, the rotation angle θ of the parking space is introduced into the regression equation. The regression branch will output (x, y, w, h, θ) to determine the multi-dimensional coordinate information of the parking space in the panoramic image. By determining the two-dimensional coordinate information of the obstacle in the panoramic image and the multi-dimensional coordinate information of the parking space in the panoramic image, effective parking space information can be obtained.

[0068] For example, as shown in Figure 5, this is a comparison of parking space detection before and after the algorithm improvement of an embodiment of the present invention. The dashed line represents the positioning box output by the target detection module. Figure 5(a) is a schematic diagram of the unimproved target detection module algorithm, and Figure 5(b) is the detection algorithm of the improved target detection module. As can be seen from Figure 5(b), in addition to detecting the width and height of the parking space, the detection of the parking space rotation angle is added, which enhances the ability to identify and detect inclined parking spaces and improves the detection capability of parking spaces.

[0069] In some embodiments, when outputting valid parking space information based on multidimensional coordinate information and two-dimensional coordinate information, the parking space boundary box is determined based on the multidimensional coordinate information; the obstacle boundary box and the center point coordinates of the obstacle are determined based on the two-dimensional coordinate information; it is determined whether the center point coordinates of the obstacle are within the parking space boundary box; if not, the boundary overlap ratio of the parking space boundary box and the obstacle boundary box is determined; if the boundary overlap ratio does not exceed a preset ratio threshold, the parking space is determined to be a valid parking space, and the valid parking space information is output.

[0070] In this embodiment, after the target detection module acquires the multidimensional coordinate information of the parking space and the two-dimensional coordinate information of the obstacle in the panoramic image, it analyzes the positional relationship between the parking space and the obstacle. Based on the multidimensional coordinate information of the parking space in the panoramic image, it determines the parking space bounding box; based on the two-dimensional coordinate information of the obstacle in the panoramic image, it determines the obstacle bounding box and the center point coordinates of the obstacle, and determines whether the center point coordinates of the obstacle are within the parking space bounding box. If the center point coordinates of the obstacle are within the parking space bounding box, it is considered that an obstacle exists within the parking space bounding box, and the parking space is considered invalid. If the center point coordinates of the obstacle are not within the parking space bounding box, the parking space bounding box and the obstacle are... The ratio of the overlapping area between the boundary of the obstacle's bounding box and the parking space's bounding box is compared with a preset ratio threshold. If the ratio is greater than the preset ratio threshold, it is considered that there is an obstacle in the parking space, and the vehicle cannot successfully park in that space, thus the parking space is determined to be invalid. If the ratio is less than the preset ratio threshold, it is considered that there is no obstacle in the parking space, or that the existing obstacle will not affect the vehicle's parking, thus the parking space is determined to be valid. By determining the relationship between the overlap ratio of the boundary of the parking space's bounding box and the obstacle's bounding box and the preset ratio threshold, the problem of vehicle parking failure due to the inability to detect obstacle information in the parking space can be avoided, thereby improving the success rate of parking.

[0071] In some embodiments, when processing a panoramic image to obtain parking line information, feature vectors are extracted from the panoramic image to determine the set of parking line pixels in the panoramic image; and parking line information is obtained based on the set of parking line pixels.

[0072] In this embodiment, after acquiring a panoramic image of the driving range of the vehicle to be parked, the convolutional neural network module uses a convolution algorithm to extract feature vectors from the panoramic image to obtain feature vectors, and sends the extracted feature vectors to the parking line detection module so that the parking line detection module can detect the parking line information.

[0073] After receiving the feature vector, the parking line detection module classifies the feature vector, that is, it determines whether the pixels in the panoramic image belong to the parking line. If the pixel is determined to belong to the parking line information, it iterates through all the pixels to obtain the parking line pixel set, and uses a mask to mark the parking line in the panoramic image according to the parking line pixel set, thereby increasing the priority of the parking line to detect the angle distance between the sides of the vehicle and the parking line, preventing the vehicle from crossing the line or tilting after parking, thus preventing irregular parking.

[0074] In some embodiments, when processing a panoramic image to obtain a drivable area, feature vectors are extracted from the panoramic image, and the pixels of the drivable area are determined in the panoramic image; the drivable area is determined based on the pixels of the drivable area.

[0075] In this embodiment, after acquiring a panoramic image of the driving range of the vehicle to be parked, the convolutional neural network module uses a convolution algorithm to extract feature vectors from the panoramic image to obtain feature vectors, and sends the extracted feature vectors to the drivable area detection module so that the drivable area detection module can detect the drivable area of ​​the road surface.

[0076] After receiving the feature vector, the drivable area detection module classifies the feature vector, determining whether pixels in the panoramic image belong to drivable areas, such as obstacles, vehicles, and pedestrians. If a pixel is determined to belong to a drivable area, it is marked and enclosed in a closed loop; this enclosed area is the drivable area, accessible to the vehicle. This drivable area detection is part of the semantic segmentation model, outputting drivable area information for autonomous parking to further plan paths and avoid obstacles, improving parking efficiency and preventing multiple instances of "clamping" during automatic or assisted parking.

[0077] In some embodiments, when determining the target parking space coordinate information based on the valid parking space information and the parking space line information, the area and angle of the parking space boundary box of the valid parking space are corrected based on the parking space line information to obtain the target parking space coordinate information.

[0078] In this embodiment, after obtaining valid parking space information through the target detection module and parking line information through the parking line detection module, the valid parking space information and parking line information are fused together and mapped onto the panoramic image to display the parking space boundary box, obstacle boundary box, and parking line information of the valid parking space. Based on the detected parking line information, the distance and angle between the vehicle's two sides and the parking line are calculated, and it is analyzed whether the vehicle has deviated or crossed the line. The size and angle information of the valid parking space boundary box are continuously corrected to obtain more accurate target parking space coordinate information, so that the vehicle can be corrected in real time according to the target parking space coordinate information when parking.

[0079] In some embodiments, when determining a target parking path based on a drivable area, multiple parking paths are determined within the drivable area; and the target parking path is determined based on the multiple parking paths.

[0080] In this embodiment, after the drivable area detection module determines the drivable area, it uses the Monte Carlo Tree Search (MCTS) algorithm to plan the parking path. That is, within the drivable area, the vehicle drivable area detection module can plan the parking route by combining obstacle information such as obstacles, pedestrians and parked vehicles, and generate multiple parking paths in sequence. The target parking path is then determined based on the multiple parking paths.

[0081] In some embodiments, when determining multiple parking paths in a drivable area, the current location information of the parked vehicle is obtained; multiple parking paths are determined based on the current location information and the coordinate information of the target parking space.

[0082] In this embodiment, the current location information of the parked vehicle is obtained, and the current location of the parked vehicle is taken as the starting point; the location information of the target parking space is obtained, and the location of the target parking space is taken as the target point. With the information of obstacles, drivable areas, parking lines, pedestrians and parked vehicles as constraints, starting from the current location of the parked vehicle, avoiding obstacles, parking lines, pedestrians and parked vehicles, the vehicle reaches the location of the target parking space, and multiple sets of parking routes are generated in sequence to determine multiple parking paths.

[0083] In some embodiments, when determining a target parking path based on multiple parking paths, a set of probability values ​​for successful parking on the multiple parking paths is obtained, and the path with the highest probability value is taken as the target parking path.

[0084] In this embodiment, a set of parking success probability values ​​corresponding to multiple parking route sets is generated sequentially. The route with the highest parking success probability is selected from the multiple parking routes as the target parking route, and the target parking route is sent to the vehicle. During the parking process, the target parking route is optimized in real time based on the vehicle's position status and the parking space position, reducing the number of times the vehicle traverses the parking space and the parking time, thereby significantly improving the parking success rate.

[0085] The following is combined Figure 6 The parking control method of this invention will be illustrated by example.

[0086] After acquiring a panoramic image of the driving range of the vehicle to be parked, the multi-task deep learning module processes the input panoramic image. Specifically, the convolutional neural network module uses a convolutional algorithm to extract feature vectors from the panoramic image, and shares these extracted feature vectors with the target detection module, parking line detection module, and drivable area detection module. The target detection module obtains valid parking space information, the parking line detection module obtains parking line information, and the drivable area detection module obtains drivable area information. The valid parking space information, parking line information, and drivable area information are then fused to determine the target parking space coordinates and the target parking path.

[0087] like Figure 7 The diagram shown is a flowchart of a parking control method according to an embodiment of the present invention. The parking control method includes at least steps S21-S23.

[0088] Step S21: Obtain a panoramic image of the driving range of the vehicle to be parked.

[0089] Step S22: Process the panoramic image based on a multi-task deep learning model to obtain effective parking space information, parking line information, and drivable area.

[0090] Step S23: Determine the target parking space coordinates based on the valid parking space information and parking line information, and determine the target parking path based on the drivable area.

[0091] According to the parking control method of this invention, when a vehicle is parking, the panoramic image is processed to obtain effective parking space information, parking line information, and drivable area. The parking line information and drivable area are taken into account to optimize the parking space coordinate information and parking path, resulting in target parking space coordinate information and target parking path. This avoids parking position deviations, makes the parking path more reasonable, reduces the number of times parking spaces need to be traversed, and improves parking efficiency. Furthermore, when determining the above information, a multi-task deep learning model is used to process the panoramic image. Compared to a combination of two deep learning models, this method has lower computational requirements, more accurate and simpler feature fusion, and eliminates the need for double detection of a single image, increasing detection speed. This reduces parking costs while improving user experience.

[0092] The parking control device according to an embodiment of the present invention is described below.

[0093] The parking control device of this invention includes: a processor, a memory, and a parking control program stored in the memory and executable on the processor. When the parking control program is executed by the processor, it implements the parking control method as described in the above embodiments.

[0094] According to an embodiment of the present invention, the parking control device processes a panoramic image when a vehicle is parking to obtain effective parking space information, parking line information, and drivable area. Taking the parking line information and drivable area into account, the device optimizes the parking space coordinate information and parking path to obtain the target parking space coordinate information and target parking path. This avoids parking position deviations, makes the parking path more reasonable, reduces the number of times vehicles need to maneuver in parking spaces, and increases parking efficiency. Furthermore, when determining the above information, a multi-task deep learning model is used to process the panoramic image. Compared to a combination of two deep learning models, this method has lower computational requirements, more accurate and simpler feature fusion, and eliminates the need for double detection of a single image, increasing detection speed. This reduces parking costs while improving user experience.

[0095] The following is for reference. Figure 8A vehicle described in an embodiment of the present invention.

[0096] like Figure 8 As shown, the vehicle 3 in this embodiment of the invention includes the parking control device 2 as described in the above embodiment.

[0097] According to the embodiment of the present invention, when the vehicle 3 is parking, the panoramic image is processed to obtain effective parking space information, parking line information, and drivable area. The parking line information and drivable area are taken into account to optimize the parking space coordinate information and parking path, thereby obtaining the target parking space coordinate information and target parking path. This avoids parking position deviation and makes the parking path more reasonable, reducing the number of times parking spaces need to be traversed, resulting in high parking efficiency. When determining the above information, a multi-task deep learning model is used to process the panoramic image. Compared with the combination of two deep learning models, it has lower computational requirements, more accurate and simpler feature fusion, and does not require two detections on a single image, increasing detection speed. This reduces parking costs while improving user experience.

[0098] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0099] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0100] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A parking control method, characterized in that, include: Acquire a panoramic image of the driving range of the vehicle waiting to be parked; The panoramic image is processed to obtain valid parking space information, parking line information, and drivable area; The panoramic image is processed by a single model using multi-task deep learning to handle multiple tasks. The target parking space coordinates are determined based on the available parking space information and the parking space line information. Determine the target parking route based on the drivable area; The vehicle to be parked is controlled to park based on the target parking space coordinates and the target parking path. The panoramic image is processed to obtain valid parking space information, including: After extracting feature vectors from the panoramic image, the multidimensional coordinate information of the parking space and the two-dimensional coordinate information of the obstacle in the panoramic image are determined. The multidimensional coordinate information includes the rotation angle of the parking space. The effective parking space information is obtained based on the multi-dimensional coordinate information and the two-dimensional coordinate information.

2. The parking control method according to claim 1, characterized in that, The effective parking space information is output based on the multi-dimensional coordinate information and the two-dimensional coordinate information, including: The parking space boundary box is determined based on the multidimensional coordinate information; The obstacle boundary frame and the center point coordinates of the obstacle are determined based on the two-dimensional coordinate information. Determine whether the center point coordinates of the obstacle are within the parking space boundary frame of the parking space; If not, then determine the overlap ratio of the boundaries of the parking space boundary frame and the obstacle boundary frame; When the overlap ratio of the boundaries does not exceed a preset ratio threshold, the parking space is determined to be a valid parking space, and the valid parking space information is output.

3. The parking control method according to claim 1, characterized in that, Processing the panoramic image to obtain the parking line information includes: After extracting feature vectors from the panoramic image, the set of parking line pixels is determined in the panoramic image; The parking space information is obtained based on the set of pixels of the parking space lines.

4. The parking control method according to claim 1, characterized in that, The panoramic image is processed to obtain the drivable area, including: After extracting feature vectors from the panoramic image, the pixels of the drivable area are determined in the panoramic image. The drivable area is determined based on the pixels of the drivable area.

5. The parking control method according to claim 3, characterized in that, Determining the target parking space coordinates based on the available parking space information and the parking space line information includes: The area and angle of the boundary frame of the effective parking space are corrected based on the parking space line information to obtain the coordinate information of the target parking space.

6. The parking control method according to claim 1, characterized in that, Determining the target parking route based on the drivable area includes: Multiple parking routes are determined within the drivable area; The target parking path is determined based on multiple parking paths.

7. The parking control method according to claim 6, characterized in that, Multiple parking routes are determined within the drivable area, including: Obtain the current location information of the parked vehicle; Multiple parking paths are determined based on the current location information and the target parking space coordinates.

8. The parking control method according to claim 6, characterized in that, Determining the target parking path based on multiple parking paths includes: Obtain a set of probability values ​​for successful parking along multiple parking paths, and select the path with the highest probability value as the target parking path.

9. A parking control device, characterized in that, include: A processor, a memory, and a parking control program stored in the memory and executable on the processor, wherein the parking control program, when executed by the processor, implements the parking control method as described in any one of claims 1-8.

10. A vehicle, characterized in that, include: The parking control device as described in claim 9.

Images