Method, device, vehicle and computer-readable storage medium for constructing parking map

Abstract

The present invention discloses a method, device, vehicle and computer-readable storage medium for constructing a parking map, the method comprising: obtaining a bird's-eye view of a vehicle, the bird's-eye view including environmental information within a preset range around the vehicle; determining parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle according to the bird's-eye view; and obtaining a parking map according to the parking space information and the drivable area. The present invention can avoid the problem of positioning loss in constructing a parking map, and does not require a large number of calculations, and the vehicle does not need to provide high computing power support, thereby improving the parking accuracy of the vehicle and reducing the amount of calculation.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a method, device, vehicle and computer-readable storage medium for constructing a parking map. Background Art

[0002] In related technologies, the vision-based SLAM (Simulation Localization and Mapping) method is usually used in automatic parking systems. However, due to the dark light environment and a large number of repeated textures in the parking garage, it is difficult for this method to accurately detect and match feature points, resulting in positioning loss. Secondly, the front-end visual odometer in this method needs to perform a large number of feature point detection and matching operations. These operations have high requirements for computing resources and consume a lot of computing resources. At the same time, it also requires the vehicle to provide high computing power support. If sufficient computing power support cannot be achieved, it may affect the real-time and accuracy of the calculation. Summary of the invention

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

[0004] To this end, an object of the present invention is to propose a method for constructing a parking map, which can avoid the problem of positioning loss in constructing a parking map, and does not require a large number of calculations, and the vehicle does not need to provide high computing power support, thereby improving the vehicle parking accuracy and reducing the amount of calculation.

[0005] To this end, a second object of the present invention is to provide a device for constructing a parking map.

[0006] To this end, a third object of the present invention is to provide a vehicle.

[0007] To this end, a fourth object of the present invention is to provide a computer-readable storage medium.

[0008] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present invention discloses a method for constructing a parking map, comprising: obtaining a bird's-eye view of a vehicle, the bird's-eye view including environmental information within a preset range around the vehicle; determining parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle according to the bird's-eye view; and obtaining a parking map according to the parking space information and the drivable area.

[0009] According to the method for constructing a parking map of an embodiment of the present invention, firstly, a bird's-eye view of the vehicle is obtained. The bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle, and then the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle can be determined according to the bird's-eye view, and a parking map is constructed according to the parking space information and the drivable area, so as to facilitate automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and the interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0010] In addition, the method for constructing a parking map according to the above embodiment of the present invention may also have the following additional technical features:

[0011] In some embodiments, obtaining a parking map based on the parking space information and the drivable area includes: acquiring track information of the vehicle; constructing a parking space map based on the track information and the parking space information; constructing a drivable area map based on the track information and the drivable area; and obtaining the parking map based on the parking space map and the drivable area map.

[0012] In some embodiments, determining the parking space information of available parking spaces around the vehicle and the drivable area of ​​the vehicle based on the bird's-eye view includes: using the bird's-eye view as input to a trained multi-task deep learning model to obtain the parking space information and the drivable area output by the multi-task deep learning model.

[0013] In some embodiments, the multi-task deep learning model includes a parking space information detection sub-model and a drivable area detection sub-model, and the method of using the bird's-eye view as the input of the trained multi-task deep learning model to obtain the parking space information and the drivable area output by the multi-task deep learning model includes: using the bird's-eye view as the input of the parking space information detection sub-model to obtain the parking space information output by the parking space information detection sub-model, wherein the parking space information includes parking space corner points; using the bird's-eye view as the input of the drivable area detection sub-model to obtain the drivable area output by the drivable area detection sub-model.

[0014] In some embodiments, before using the bird's-eye view as the input of a trained multi-task deep learning model, it also includes: when it is determined that the bird's-eye view meets a preset output condition, using the bird's-eye view as the input of the trained multi-task deep learning model; when it is determined that the bird's-eye view does not meet the preset output condition, reacquiring the bird's-eye view.

[0015] In some embodiments, determining whether the bird's-eye view meets a preset output condition includes: when it is determined that a set parameter used to characterize image quality in the bird's-eye view meets a preset parameter threshold, determining that the bird's-eye view meets the preset output condition.

[0016] In some embodiments, obtaining the track information of the vehicle includes: obtaining the vehicle speed and heading angle of the vehicle; and obtaining the track information according to the vehicle speed and heading angle.

[0017] In some embodiments, constructing a parking map based on the track information and the parking space information includes: correcting a track error of the track information based on the parking space information to obtain corrected target track information; and constructing a parking map based on the target track information.

[0018] In some embodiments, the correcting the track error of the track information according to the parking space information to obtain the corrected target track information includes: obtaining a first key frame, wherein when the parking space information of a new available parking space is detected based on the bird's-eye view, it is determined that the new available parking space is detected, and the first key frame is formed, and the first key frame includes the parking space information of the new available parking space and the newly added drivable area; constructing a second key frame according to the proportion of the new available parking space appearing in the first key frame; determining the correspondence between the first key frame and the second key frame, wherein the correspondence includes the intersection-and-union ratio between the parking spaces in the first key frame and the parking spaces in the second key frame; according to the correspondence, correcting the track error of the track information by a preset position correction algorithm to obtain the corrected target track information, wherein the track error includes the track motion error generated during the movement of the vehicle and the track calculation error generated during the track calculation.

[0019] In some embodiments, constructing a parking space map according to the target track information includes: fusing the target track information with the parking space information in the second key frame to form the parking space map.

[0020] In some embodiments, after forming the parking map, the method further includes: correcting errors in the parking information of the newly added available parking spaces in the parking map by using a preset filtering algorithm to obtain a corrected parking map.

[0021] In some embodiments, constructing a drivable area map according to the track information and the drivable area includes: fusing the target track information and the drivable area based on a preset map fusion algorithm to obtain the drivable area map.

[0022] In some embodiments, the target track information and the drivable area are fused based on a preset map fusion algorithm, including: obtaining obstacle information detected at multiple different times during the movement of the vehicle; based on the preset map fusion algorithm, the obstacle information detected at multiple different times is fused, and the real position and size of the obstacle are restored on a two-dimensional grid map to obtain the drivable area map.

[0023] In some embodiments, obtaining the parking map according to the parking space map and the drivable area map includes: merging the parking space map and the drivable area map to obtain the parking map.

[0024] In some embodiments, obtaining a bird's-eye view of the vehicle includes: obtaining a plurality of initial pictures around the vehicle; and splicing the plurality of initial pictures to obtain the bird's-eye view.

[0025] To achieve the above-mentioned purpose, an embodiment of the second aspect of the present invention proposes a device for constructing a parking map, comprising: an acquisition module, used to acquire a bird's-eye view of a vehicle, wherein the bird's-eye view includes environmental information within a preset range around the vehicle; a first processing module, used to determine parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle based on the bird's-eye view; and a second processing module, used to obtain a parking map based on the parking space information and the drivable area.

[0026] According to the device for constructing a parking map of the embodiment of the present invention, the acquisition module first obtains a bird's-eye view of the vehicle, and the bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle. The first processing module can then determine the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle according to the bird's-eye view, and the second processing module constructs a parking map according to the parking space information and the drivable area, thereby facilitating automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0027] In order to achieve the above-mentioned purpose, an embodiment of the third aspect of the present invention proposes a vehicle, comprising the device for constructing a parking map as described in the embodiment of the second aspect above; or, a processor, a memory, and a program for constructing a parking map stored in the memory and executable on the processor, wherein the program for constructing a parking map, when executed by the processor, implements the method for constructing a parking map as described in the embodiment of the first aspect of the present invention.

[0028] According to the vehicle of the embodiment of the present invention, a bird's-eye view of the vehicle is first obtained. The bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle, and then the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle can be determined according to the bird's-eye view. A parking map is constructed according to the parking space information and the drivable area, thereby facilitating automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0029] In order to achieve the above-mentioned purpose, an embodiment of the fourth aspect of the present invention proposes a computer-readable storage medium, on which a program for constructing a parking map is stored. When the program for constructing a parking map is executed by a processor, the method for constructing a parking map described in the embodiment of the first aspect of the present invention is implemented.

[0030] According to the computer-readable storage medium of the embodiment of the present invention, when the program for constructing a parking map stored thereon is executed by the processor, the bird's-eye view of the vehicle is first obtained. The bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle, and then the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle can be determined according to the bird's-eye view, and the parking map is constructed according to the parking space information and the drivable area, so as to facilitate automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and the interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0031] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0033] Figure 1 is a flowchart of a method for constructing a parking map according to an embodiment of the present invention;

[0034] Figure 2 is a schematic diagram of track information of a vehicle according to an embodiment of the present invention;

[0035] Figure 3 is a schematic diagram of a map fusion algorithm according to an embodiment of the present invention;

[0036] Figure 4 is a schematic diagram of a parking map according to an embodiment of the present invention;

[0037] Figure 5 is a structural block diagram of an apparatus for constructing a parking map according to an embodiment of the present invention;

[0038] Figure 6 is a structural block diagram of a vehicle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] Embodiments of the present invention are described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. Embodiments of the present invention are described in detail below.

[0040] Reference below Figure 1-Figure 4 A method for constructing a parking map according to an embodiment of the present invention is described.

[0041] Figure 1 FIG. 1 is a flow chart of a method for constructing a parking map according to an embodiment of the present invention. Figure 1 As shown, the method for constructing a parking map of the present invention at least includes steps S1 to S3.

[0042] Step S1, obtaining a bird's-eye view of the vehicle, where the bird's-eye view includes environmental information within a preset range around the vehicle.

[0043] In an embodiment, a bird's-eye view of the vehicle can be obtained through cameras, which include, for example, but are not limited to four fish-eye cameras. Fish-eye cameras have advantages such as ultra-wide viewing angles and are suitable for capturing panoramic images. The initial images captured by the fish-eye cameras are used to obtain a bird's-eye view through image stitching technology. The obtained bird's-eye view will present environmental information within a preset range around the vehicle, thereby facilitating the driver to locate the vehicle according to the bird's-eye view and improving the accuracy of determining parking space information and drivable areas.

[0044] Step S2, determining parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle according to the bird's-eye view.

[0045] The parking space information includes, for example, parking space corner points, which refer to intersections in a parking garage used to identify the boundaries of a specific parking space; the drivable area refers to a road surface where there are no obstacles and the vehicle can be driven safely during parking.

[0046] In an embodiment, after obtaining a bird's-eye view, the obtained bird's-eye view is processed to determine parking space information of available parking spaces around the vehicle and the drivable area of ​​the vehicle. For example, the bird's-eye view is input into a multi-task deep learning model and parsed by the multi-task deep learning model. The parsing process includes, for example, segmenting the drivable area and identifying parking space corner points. After the parsing is completed, the parking space information of the available parking spaces and the drivable area of ​​the vehicle can be obtained according to the parsing results of the multi-task deep learning model, thereby improving the accuracy of constructing the parking map.

[0047] Step S3, obtaining a parking map according to the parking space information and the drivable area.

[0048] In the embodiment, for example, a camera is used to obtain parking space information within a preset range around the vehicle, that is, the camera can collect the corner points of the vehicle's parking space, and also detect and determine the drivable area within the preset range around the vehicle, that is, the area without obstacles and where the vehicle can pass safely. After obtaining the parking space information and the drivable area, the parking space information and the drivable area are integrated to construct a parking map. The parking map not only intuitively displays the distribution of available parking spaces around the vehicle, but also clearly displays the paths and areas where the vehicle can drive safely, thereby providing the driver with comprehensive parking guidance, helping the driver to quickly find a suitable parking space and plan a safe and efficient parking path, greatly improving the parking positioning accuracy.

[0049] According to the method for constructing a parking map of an embodiment of the present invention, firstly, a bird's-eye view of the vehicle is obtained. The bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle, and then the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle can be determined according to the bird's-eye view, and a parking map is constructed according to the parking space information and the drivable area, so as to facilitate automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and the interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0050] In one embodiment of the present invention, a parking map is obtained based on parking space information and a drivable area, including: obtaining track information of a vehicle; constructing a parking space map based on the track information and the parking space information; constructing a drivable area map based on the track information and the drivable area; and obtaining a parking map based on the parking space map and the drivable area map.

[0051] The track information includes the vehicle's travel distance and heading angle, which can also be understood as the travel path or posture change.

[0052] In an embodiment, wheel speed sensors are installed on the four wheels of the vehicle. For example, the vehicle speed and heading angle are obtained through the wheel speed sensors. The vehicle track information is obtained by calculation based on the obtained speed and heading angle, thereby providing a driving path for the vehicle to park and improve the accuracy of parking.

[0053] Among them, the track information records the vehicle's driving path and provides the driver with the actual driving track of the vehicle in the parking garage. At the same time, the parking space information includes the parking corner points of each parking space. These parking corner points are used as feature points and adjusted to accurately reflect the actual parking space location. Therefore, combining the track information and the parking space information can obtain a more accurate parking space location, thereby building a reliable parking space map.

[0054] Next, based on the track information, it is corrected to obtain corrected target track information, and the corrected target track information is merged with the drivable area to obtain a drivable area map.

[0055] The parking space map and the drivable area map are then merged to obtain a parking map that includes both parking space information and drivable area. The track information and parking space information presented by the parking map are more accurate. During the movement of the vehicle, it can help the driver find a reliable parking location and clearly understand the drivable area, which facilitates the driver to park accurately and improves parking accuracy.

[0056] In one embodiment of the present invention, parking space information of available parking spaces around a vehicle and a drivable area of ​​the vehicle are determined based on a bird's-eye view, including: using the bird's-eye view as an input to a trained multi-task deep learning model to obtain the parking space information and drivable area output by the multi-task deep learning model.

[0057] Among them, the multi-task deep learning model is a multi-task model that can handle multiple different tasks at the same time, for example, it can perform semantic segmentation and target detection tasks at the same time.

[0058] In an embodiment, the acquired bird's-eye view is taken as input, and the trained multi-task deep learning model parses the received bird's-eye view. During the parsing process, the trained multi-task deep learning model can use its own learning characteristics, such as convolutional neural networks, to extract key features from the bird's-eye view, and combine and classify these features. After the parsing is completed, the trained multi-task deep learning model can output parking space information and drivable areas, wherein the parking space information includes, for example, parking space corner points, and the parking space information may also include parking space size, location and direction, etc. The drivable area is the area range where the vehicle can drive safely.

[0059] By adopting a trained multi-task deep learning model, not only the accuracy of parking space information and drivable area acquisition is improved, but also the parallel processing of two tasks is achieved.

[0060] In one embodiment of the present invention, the multi-task deep learning model includes a parking space information detection sub-model and a drivable area detection sub-model. The bird's-eye view is used as the input of the trained multi-task deep learning model to obtain the parking space information and drivable area output by the multi-task deep learning model, including: using the bird's-eye view as the input of the parking space information detection sub-model to obtain the parking space information output by the parking space information detection sub-model, and the parking space information includes parking space corner points; using the bird's-eye view as the input of the drivable area detection sub-model to obtain the drivable area output by the drivable area detection sub-model.

[0061] In an embodiment, the bird's-eye view image is input into a trained multi-task deep learning model, and is parsed separately in two sub-models of the multi-task deep learning model, wherein the multi-task deep learning model uses lightweight convolutional neural networks such as MobileNet and ShuffleNet to identify parking corners and drivable areas of vehicles.

[0062] Specifically, the bird's-eye view is input into the parking space information detection sub-model of the multi-task deep learning model. The preset target recognition algorithm in the parking space information detection sub-model can be used to identify the parking corner points of the available parking spaces in the bird's-eye view, such as SSD (Single Shot MultiBox Detector) or YOLO (YouOnly Look Once, a target detection algorithm based on a deep neural network). These target recognition algorithms can accurately locate and identify the targets in the bird's-eye view, that is, the parking corner points, and give the position information in the bird's-eye view, that is, identify the available parking spaces within a preset range around the vehicle, thereby obtaining the parking space information output by the parking space information detection sub-model. The use of the target recognition algorithm can achieve fast calculation while ensuring the identification of parking corner points.

[0063] On the other hand, in the drivable area detection sub-model, the drivable area of ​​the vehicle in the bird's-eye view is identified based on a preset semantic segmentation algorithm, such as using semantic segmentation algorithms such as UNet (U-shaped network) or Deeplab (deep label). These algorithms can classify each pixel in the bird's-eye view into a specific category, thereby distinguishing between drivable areas and non-drivable areas. That is, the semantic segmentation algorithm can generate a pixel-level segmentation result. Based on the segmentation result, the drivable area output by the drivable area detection sub-model can be accurately obtained.

[0064] Among them, both the target recognition algorithm and the semantic segmentation algorithm use lightweight convolutional neural networks to reduce computational complexity, improve parsing speed, and meet real-time requirements.

[0065] In one embodiment of the present invention, before using the bird's-eye view as the input of the trained multi-task deep learning model, it also includes: when it is determined that the bird's-eye view meets the preset output conditions, using the bird's-eye view as the input of the trained multi-task deep learning model; when it is determined that the bird's-eye view does not meet the preset output conditions, re-acquiring the bird's-eye view.

[0066] In an embodiment, before using the bird's-eye view as the input of the trained multi-task deep learning model, it is necessary to further determine whether the acquired bird's-eye view meets the preset output conditions. If it is determined that the bird's-eye view meets the preset output conditions, the bird's-eye view is used as the input of the trained multi-task deep learning model; if it is determined that the bird's-eye view does not meet the preset output conditions, the bird's-eye view is reacquired, for example, by recapturing the image using a camera. This process is repeated until a bird's-eye view that meets the preset output conditions is acquired. This ensures the quality of the bird's-eye view input into the multi-task deep learning model and improves the accuracy and reliability of the multi-task deep learning model analysis.

[0067] In one embodiment of the present invention, determining whether the bird's-eye view image satisfies a preset output condition includes: when it is determined that a set parameter used to characterize image quality in the bird's-eye view image satisfies a preset parameter threshold, determining that the bird's-eye view image satisfies the preset output condition.

[0068] Among them, the set parameters used to characterize the image quality include, for example, image clarity, resolution, etc.; the preset parameter threshold is set according to demand and is used to determine whether the acquired bird's-eye view meets the preset output conditions.

[0069] In an embodiment, the set parameter used to characterize the image quality is compared with a preset parameter threshold, and if the set parameter meets the preset parameter threshold, it can be determined that the bird's-eye view image meets the preset output condition.

[0070] Specifically, when the clarity and resolution of the acquired bird's-eye view reach certain standards, that is, these set parameters all meet the preset parameter thresholds, the acquired bird's-eye view can be considered as an image that meets the image quality standards and has sufficient clarity and resolution. It can be used as the input of a multi-task deep learning model, thereby accurately identifying the parking corners and drivable areas in the image.

[0071] In one embodiment of the present invention, obtaining the track information of the vehicle includes: obtaining the vehicle speed and heading angle of the vehicle; and obtaining the track information according to the vehicle speed and heading angle.

[0072] In an embodiment, the vehicle speed and heading angle can be obtained through the wheel speed sensor, wherein the heading angle refers to the angle between the direction of vehicle travel and a fixed direction. After the vehicle speed and heading angle are obtained, the vehicle's track information can be calculated, thereby achieving accurate positioning of the vehicle's travel path.

[0073] In one embodiment of the present invention, obtaining the vehicle speed includes: obtaining a first wheel speed corresponding to a first wheel and a second wheel speed corresponding to a second wheel of the vehicle; and obtaining the vehicle speed according to an average of the first wheel speed and the second wheel speed.

[0074] In an embodiment, in order to obtain the vehicle speed, a wheel speed sensor may be installed at the wheel, and a first wheel speed corresponding to a first wheel and a second wheel speed corresponding to a second wheel of the vehicle are respectively obtained through the wheel speed sensor. The first wheel speed is, for example, denoted as v l , the second wheel speed is denoted as v r , then calculate the first wheel speed v l and the second wheel speed v r The mean, for example, is denoted by v vehicle , the calculated mean v vehicle As the vehicle speed, , which can reduce the error and ensure the accuracy of the vehicle speed data.

[0075] In one embodiment of the present invention, obtaining the vehicle heading angle includes: obtaining the difference in movement distance between a first wheel and a second wheel of the vehicle within a target time, and the wheelbase between the first wheel and the second wheel, wherein the target time is the time taken for the first wheel or the second wheel to make one rotation; and obtaining the heading angle according to the ratio of the movement distance difference and the wheelbase.

[0076] In an embodiment, the difference in movement distances of a first wheel and a second wheel of a vehicle within a target time is obtained, and the movement distance difference is recorded as d, wherein the target time refers to the time required for any one of the first wheel and the second wheel to complete a full rotation first. For example, if the first wheel completes a full rotation first, the time taken by the first wheel to complete one rotation is taken as the target time. Similarly, if the second wheel completes a full rotation first, the time taken by the second wheel to complete one rotation is taken as the target time.

[0077] At the same time, the wheelbase between the first wheel and the second wheel is obtained, for example, the distance between the centers of the two wheels is recorded as t. Then, the heading angle of the vehicle is calculated based on the ratio of the movement distance difference d of the two wheels within the target time and the wheelbase t and recorded as ,Right now .

[0078] In one embodiment of the present invention, obtaining the difference in movement distance between a first wheel and a second wheel of a vehicle within a target time includes: obtaining a first wheel speed corresponding to the first wheel and a second wheel speed corresponding to the second wheel, and a first difference in time taken for the first wheel and the second wheel to respectively rotate one circle; determining a second difference between the first wheel speed and the second wheel speed; and obtaining the movement distance difference according to the product of the first difference and the second difference.

[0079] In the embodiment, the first wheel speed v corresponding to the first wheel is obtained by a wheel speed sensor. l The second wheel speed v corresponding to the second wheel r , since the speeds of the first wheel and the second wheel are different when the vehicle is turning, the time required for the first wheel and the second wheel to complete one rotation is also different. Then, after obtaining the time required for the first wheel and the second wheel to complete one rotation, the first difference in the time taken for the first wheel and the second wheel to complete one rotation is calculated, for example, recorded as Δt, and at the same time, the first wheel speed v is determined l and the second wheel speed v r The second difference between .

[0080] Specifically, every time the wheel rotates one circle, the wheel speed sensor will generate a pulse signal point with a fixed frequency. According to the wheel size and the distance between each two pulse signals, and then according to the time difference between the two pulse signals, the first wheel speed v can be calculated. land the second wheel speed v r , therefore, the first wheel speed v l and the second wheel speed v r The second difference between l -v r ).

[0081] Finally, the first difference Δt and the second difference (v l -v r ) can be multiplied to obtain the first difference Δt and the second difference (v l -v r ), we get (v l -v r ) Δt, and use the product as the moving distance difference d, that is, d= (v l -v r ) Δt, thereby obtaining the movement distance difference d.

[0082] In summary, if Figure 2 The schematic diagram of the vehicle's track information is shown in FIG. The wheels are turned along the center point O. The vehicle speed can be calculated by the average of the first wheel speed and the second wheel speed. The vehicle heading angle can be obtained by the ratio of the difference in movement distance between the first wheel and the second wheel and the wheelbase. Therefore, the track information can be obtained based on the obtained vehicle speed and heading angle. The calculation is shown in formula (1), that is:

[0083] ;

[0084] Among them, v vehicle is the vehicle speed, v l is the first wheel speed, v r is the second wheel speed, d is the moving distance difference, Δt is the first difference, (v l -v r ) is the second difference, is the heading angle, and t is the wheelbase.

[0085] In one embodiment of the present invention, the first wheel is the left front wheel, and the second wheel is the right front wheel; or, the first wheel is the left rear wheel, and the second wheel is the right rear wheel.

[0086] In a specific embodiment, when the first wheel is the left front wheel and the second wheel is the right front wheel, by measuring the wheel speeds of the two wheels, since the wheel speeds of the first wheel and the second wheel are different when the vehicle is turning, the paths traveled are different due to the different wheel speeds. In this way, the travel distance between the first wheel and the second wheel will have a certain difference, and the movement distance difference d between the first wheel and the second wheel when the vehicle is turning can be calculated based on the distance difference.

[0087] Alternatively, the first wheel is the left rear wheel and the second wheel is the right rear wheel. According to the wheel speed difference between the two wheels, i.e., the second difference (v l -v r ), the difference d between the moving distances of the two wheels can be calculated, and the heading angle of the wheels can be further calculated. , thereby achieving accurate judgment of the vehicle's driving direction and improving positioning accuracy.

[0088] In one embodiment of the present invention, a parking space map is constructed based on track information and parking space information, including: correcting the track error of the track information based on the parking space information to obtain corrected target track information; and constructing the parking space map based on the target track information.

[0089] In the embodiment, since the accuracy of the track information may be affected by various factors, such as temperature, dynamic load, and tire pressure, these factors may affect the deformation and motion state of the wheel, and then affect the measurement accuracy of the vehicle speed, thereby affecting the accuracy of the track information. Therefore, directly constructing a parking map through track information may cause drift in vehicle positioning and mapping. In order to improve the accuracy of constructing a parking map, after obtaining the track information, the track error of the track information is corrected according to the parking information, so that the track error in the track information can be corrected to obtain the corrected target track information. The target track information can more accurately reflect the actual motion trajectory of the vehicle. Then, a parking map is constructed based on the target track information, which can improve the accuracy of positioning and mapping.

[0090] In one embodiment of the present invention, the track error of the track information is corrected according to the parking space information to obtain the corrected target track information, including: obtaining a first key frame, wherein when the parking space information of a new available parking space is detected based on the bird's-eye view, it is determined that the new available parking space is detected, and a first key frame is formed, the first key frame including the parking space information of the new available parking space and the newly added drivable area; constructing a second key frame according to the proportion of the new available parking space appearing in the first key frame; determining the correspondence between the first key frame and the second key frame, wherein the correspondence includes the intersection-and-union ratio between the parking spaces in the first key frame and the parking spaces in the second key frame; according to the correspondence, the track error of the track information is corrected by a preset position correction algorithm to obtain the corrected target track information, wherein the track error includes the track motion error generated during the vehicle movement and the track calculation error generated during the track calculation.

[0091] Specifically, when constructing a parking map, when parking information of new available parking spaces is detected in the acquired bird's-eye view, for example, when a new parking corner point appears, it is determined that a new available parking space is detected. Based on the new parking corner point, it can be determined that there is parking information of new available parking spaces, and the corresponding newly added drivable area will also be determined. At this time, based on the parking information of the new available parking spaces and the newly added drivable area, a first key frame is obtained, and then a second key frame is constructed based on the proportion of new available parking spaces appearing in the first key frame. Then, the corresponding relationship between the first key frame and the second key frame is determined. Based on the corresponding relationship, the track error of the track information is corrected by a preset position correction algorithm. In this way, constructing a parking map based on the target track information can provide the driver with a clear parking space usage situation, which is conducive to improving the parking positioning accuracy.

[0092] Among them, when a new available parking space is detected based on the bird's-eye view, a first key frame will be formed. The first key frame includes the parking space information of the new available parking space and the newly added drivable area, such as the new parking corner point, and the newly added drivable area can provide a safe new parking space path for the vehicle. In this way, the vehicle can detect new available parking spaces in real time according to the changes of the vehicle during movement, thereby further updating the parking map in real time, providing the driver with accurate and real-time parking space information and drivable area, which is conducive to improving the driver's parking accuracy.

[0093] When constructing the second key frame according to the proportion of new available parking spaces in the first key frame, the parking spaces in the first key frame and the parking spaces in the second key frame are extracted, and the intersection-and-union ratio between the parking spaces in the two key frames is calculated, wherein the intersection-and-union ratio refers to the ratio of the intersection area to the union area between the parking spaces in the first key frame and the parking spaces in the second key frame. The corresponding relationship between the first key frame and the second key frame can be determined according to the intersection-and-union ratio, thereby realizing the updating of the parking space map and improving the accuracy of constructing the parking space map.

[0094] After obtaining the correspondence between the first key frame and the second key frame, the track error of the track information can be corrected by a preset position correction algorithm, wherein the track error includes the track motion error generated during the vehicle movement and the track calculation error generated during the track calculation. The preset position correction algorithm is, for example, a BA (Bundle Adjustment) algorithm. The BA algorithm will use the correspondence to correct the track error of the track information, thereby obtaining the corrected target track information. The target track information can more accurately reflect the vehicle's driving path and improve the accuracy of constructing the parking map.

[0095] In one embodiment of the present invention, constructing a parking space map according to the target track information includes: fusing the target track information with the parking space information in the second key frame to form a parking space map.

[0096] In the embodiment, after obtaining the target track information, the target track information is fused with the parking space information in the second key frame. Specifically, after correcting the track motion error generated during the vehicle movement and the track calculation error generated during the track calculation, the corrected target track information is obtained, which can be understood as the target track being the corrected driving path, and the corrected driving path is re-fused with the parking space corner point in the second key frame to form a parking space map, which can improve the accuracy of the parking space map construction and thus improve the parking accuracy.

[0097] In one embodiment of the present invention, after the parking map is constructed, the method further includes: correcting errors in the parking space information of the newly added available parking spaces in the parking map by using a preset filtering algorithm to obtain a corrected parking map.

[0098] In an embodiment, after the parking map is constructed, since there are errors in the parking space information in the second key frame when detecting frame by frame, and there are certain errors when calculating the track information, in order to eliminate the errors and ensure the accuracy of the parking map, the errors in the parking space information of the newly added available parking spaces in the parking map can be corrected by a preset filtering algorithm. The preset filtering algorithm includes, for example, Kalman filtering, so that the corrected parking space information can be obtained, thereby obtaining a corrected parking map.

[0099] In one embodiment of the present invention, a drivable area map is constructed according to track information and the drivable area, including: based on a preset map fusion algorithm, the target track information and the drivable area are fused to obtain the drivable area map.

[0100] In an embodiment, based on track information, such as initial track information, target track information can be obtained through the BA algorithm. Then, a preset map fusion algorithm is used to fuse the target track information and the drivable area to improve the accuracy of the track information, and form a unified map with the drivable area, and finally obtain a drivable area map, thereby providing the driver with an accurate drivable area for the vehicle, reflecting the vehicle's driving trajectory, helping to provide accurate path planning, and thus facilitating improving parking accuracy.

[0101] In one embodiment of the present invention, the target track information and the drivable area are fused based on a preset map fusion algorithm, including: obtaining obstacle information detected at multiple different times during the movement of the vehicle; based on a preset map fusion algorithm, fusing the obstacle information detected at multiple different times, and restoring the real position and size of the obstacle on a two-dimensional grid map to obtain a drivable area map.

[0102] Among them, two-dimensional grid is a common map representation method, which divides the map into multiple equal-sized grids (grids). Each grid represents a small area and is marked according to whether the area is drivable or has obstacles.

[0103] In an embodiment, obstacle information detected by the vehicle at different times during movement is obtained, for example, information such as the size and position of the obstacle can be detected, and a preset map fusion algorithm is used to fuse the obstacle information detected at multiple different times, wherein these moments can be multiple continuous moments or multiple discontinuous moments, and the real position and size of the obstacle are more accurately restored on a two-dimensional grid map to display the drivable area of ​​the vehicle, thereby obtaining a drivable area map.

[0104] In an embodiment, the preset map fusion algorithm adopts, for example, a method based on a Bayesian function to fuse map information by calculating the conditional probability between obstacle information and an existing drivable area.

[0105] For example, Figure 3 As shown in the figure, the vehicle detected obstacle information at three different times t0, t1 and t2, that is, the same pillar. Due to the change of viewing angle, the pillar appears as a long obstacle on the map, while the space occupied by the pillar in actual application is limited to its root (the part in the box in the figure).

[0106] The obstacle information seen at t0, t1 and t2 is fused on the two-dimensional grid map using the Bayesian function to restore the true position and size of the obstacle. The specific calculation is shown in formula (2):

[0107] ;

[0108] Among them, p is the probability that the two-dimensional grid is occupied by an obstacle, and its threshold is between 0 and 1; new is the new Bayesian probability, which indicates the ratio of the grid being occupied by obstacles and being passable at the current moment; o old is the historical cumulative value of all Bayesian probabilities of the grid; s is an empirical parameter used to prevent the probability of the vehicle from accumulating rapidly when it is stationary.

[0109] When a newly added drivable area is detected, the pixel value of the detection result (between 0 and 255) is converted into a probability value (i.e., scaled to between 0 and 1). Then, the probability distribution of the drivable area can be obtained by traversing the two-dimensional grid map composed of the detection results according to formula (2). The probability distribution of the drivable area detected in each frame is superimposed to obtain a drivable area map.

[0110] In one embodiment of the present invention, obtaining a parking map according to a parking space map and a drivable area map includes: merging the parking space map and the drivable area map to obtain a parking map.

[0111] In an embodiment, the parking space map is corrected to ensure the accuracy of the parking space information, and then the corrected parking space map and the drivable area map are merged, and the map information of the two is superimposed together to obtain a parking map, thereby providing the driver with a more reliable parking map and improving the accuracy of parking.

[0112] For example, after the vehicle moves about 64 meters, it starts to go straight from the lower right corner and turns left at the end of the road. During the driving process, a total of 18 available parking spaces are detected, and there are obstacles (such as a pillar blocking) between every three available parking spaces on the left. Figure 4 The parking map schematic diagram shown in the figure can integrate the obstacle information detected at multiple different times during the vehicle movement, and can restore the real position and size of the obstacle more accurately on the two-dimensional grid map. Even if there are vehicles parked side by side continuously on the parking map, the real position and size of the obstacle can be restored on the parking map without being affected by the viewing angle, thereby improving the parking positioning accuracy.

[0113] In one embodiment of the present invention, obtaining a bird's-eye view of a vehicle includes: obtaining a plurality of initial images around the vehicle; and splicing the plurality of initial images to obtain a bird's-eye view.

[0114] In an embodiment, multiple initial pictures around the vehicle can be obtained by using fisheye cameras arranged around the vehicle, and then the multiple initial pictures obtained are stitched together. During the stitching process, for example, the position and size of the initial pictures can be adjusted. After the stitching is completed, a bird's-eye view can be obtained to comprehensively display the environmental information within a preset range around the vehicle.

[0115] Specifically, multiple cameras are distributed around the vehicle, for example, and these cameras are installed above the front and rear bumpers and below the left and right rearview mirrors of the vehicle. These cameras work independently to obtain multiple initial pictures around the vehicle according to the one-to-one correspondence of these cameras. In the process of obtaining the initial pictures, in order to ensure that the camera can obtain more obstacle information, this can be achieved by adjusting the angle between the lens and the ground.

[0116] When adjusting the angle, for example, an angle adjustment instruction may be sent externally. After receiving the external angle adjustment instruction, the angle between the lens of at least one of the multiple cameras and the ground may be adjusted in response to the angle adjustment instruction. For example, if one of the lenses of the camera needs to be facing the ground at a certain tilt angle, an angle adjustment instruction may be issued based on the angle requirement to ensure that the lens of the camera can be facing the ground at a specific tilt angle, thereby obtaining more obstacle information on the ground, providing the vehicle with a more accurate parking map, and improving parking accuracy.

[0117] Among them, the initial picture includes, for example, a wide-angle picture, which is taken by a wide-angle lens, such as a lens of a fisheye camera. Using the wide-angle picture as the initial picture, that is, using a fisheye camera, can capture relatively wide environmental information, thereby providing more comprehensive and accurate environmental information within a preset range around the vehicle.

[0118] In a specific embodiment, for example, the captured multiple initial images are converted through an inverse perspective transformation algorithm, so that the multiple initial images are projected into a first coordinate system with the ground as a coordinate plane and the center of the rear axle of the vehicle as the coordinate origin. Then, each point in the first coordinate system is projected according to a preset bird's-eye view template to obtain a complete bird's-eye view that includes environmental information within a preset range around the vehicle.

[0119] The inverse perspective transformation algorithm is shown in formula (3), which converts the pixel coordinates in the initial image (i.e. and ) is converted into the vehicle coordinate system (i.e., the first coordinate system). Since the fisheye camera is installed at different positions of the vehicle, the position of the initial image obtained by the fisheye camera in the vehicle coordinate system is also different. Therefore, the pixel points of the initial image are converted to the vehicle coordinate system through the inverse perspective transformation algorithm, so that the position of the pixel points in the initial image in the vehicle coordinate system can be obtained.

[0120] Next, each point in the first coordinate system is projected into a preset bird's-eye view template. Each point in the first coordinate system can be projected onto a two-dimensional plane of the bird's-eye view to obtain the pixel point coordinates under the bird's-eye view, that is, bev and bev ,These coordinates can be used to draw pixels in the preset bird’s-eye view template to obtain a synthetic bird’s-eye view.

[0121] ;

[0122] in, and is the coordinate of the pixel point of the initial image; K represents the intrinsic parameter matrix of the camera; K-1 represents the inverse projection matrix of the camera, that is, the transformation matrix from the coordinates of the pixels of the initial image to the vehicle coordinate system; T is the transformation matrix, that is, the transformation matrix for converting the camera coordinates to the coordinates of the center of the rear axle of the vehicle. In other words, the pixels in the initial image can be converted to the first coordinate system with the center of the rear axle of the vehicle as the coordinate origin through the transformation matrix T; is an undetermined parameter, indicating the position of the ground plane; c and c is the position of the pixel in the initial image in the vehicle coordinate system; bev and bev is the pixel coordinates from the bird's-eye view, that is, the corresponding coordinates of the pixel points in the initial image from the bird's-eye view; K bev The projection matrix for transforming the vehicle coordinate system to the bird's-eye view coordinate system.

[0123] The second embodiment of the present invention provides a device for constructing a parking map, such as Figure 5 As shown, the device 1 for constructing a parking map includes: an acquisition module 11 , a first processing module 12 and a second processing module 13 .

[0124] Among them, the acquisition module 11 is used to obtain a bird's-eye view of the vehicle, which includes environmental information within a preset range around the vehicle; the first processing module 12 is used to determine the parking space information of available parking spaces around the vehicle and the drivable area of ​​the vehicle based on the bird's-eye view; the second processing module 13 is used to obtain a parking map based on the parking space information and the drivable area.

[0125] In one embodiment of the present invention, the second processing module 13 obtains a parking map based on the parking space information and the drivable area, including: obtaining the vehicle's track information; constructing a parking space map based on the track information and the parking space information; constructing a drivable area map based on the track information and the drivable area; and obtaining a parking map based on the parking space map and the drivable area map.

[0126] In one embodiment of the present invention, the first processing module 12 determines the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle based on the bird's-eye view, including: using the bird's-eye view as the input of a trained multi-task deep learning model to obtain the parking space information and drivable area output by the multi-task deep learning model.

[0127] In one embodiment of the present invention, the multi-task deep learning model includes a parking space information detection sub-model and a drivable area detection sub-model. The first processing module 12 uses the bird's-eye view as the input of the trained multi-task deep learning model to obtain the parking space information and drivable area output by the multi-task deep learning model, including: using the bird's-eye view as the input of the parking space information detection sub-model to obtain the parking space information output by the parking space information detection sub-model, and the parking space information includes parking space corner points; using the bird's-eye view as the input of the drivable area detection sub-model to obtain the drivable area output by the drivable area detection sub-model.

[0128] In one embodiment of the present invention, before using the bird's-eye view as the input of the trained multi-task deep learning model, the first processing module 12 is also used to: when it is determined that the bird's-eye view meets the preset output conditions, use the bird's-eye view as the input of the trained multi-task deep learning model; when it is determined that the bird's-eye view does not meet the preset output conditions, the acquisition module 11 re-acquires the bird's-eye view.

[0129] In one embodiment of the present invention, the first processing module 12 determines whether the bird's-eye view image meets the preset output condition, including: when it is determined that the set parameters used to characterize the image quality in the bird's-eye view image meet the preset parameter threshold, determining that the bird's-eye view image meets the preset output condition.

[0130] In one embodiment of the present invention, the second processing module 13 obtains the track information of the vehicle, including: obtaining the vehicle speed and heading angle of the vehicle; and obtaining the track information according to the vehicle speed and heading angle.

[0131] In one embodiment of the present invention, the second processing module 13 obtains the vehicle speed by: obtaining a first wheel speed corresponding to a first wheel and a second wheel speed corresponding to a second wheel of the vehicle; and obtaining the vehicle speed according to the average of the first wheel speed and the second wheel speed.

[0132] In one embodiment of the present invention, the second processing module 13 obtains the vehicle heading angle, including: obtaining the difference in movement distance between a first wheel and a second wheel of the vehicle within a target time, and the wheelbase between the first wheel and the second wheel, wherein the target time is the time taken for the first wheel or the second wheel to make one rotation; and obtaining the heading angle according to the ratio of the movement distance difference and the wheelbase.

[0133] In one embodiment of the present invention, the second processing module 13 obtains the difference in movement distance between a first wheel and a second wheel of the vehicle within a target time, including: obtaining a first wheel speed corresponding to the first wheel and a second wheel speed corresponding to the second wheel, and a first difference in time taken for the first wheel and the second wheel to rotate one circle; determining a second difference between the first wheel speed and the second wheel speed; and obtaining the movement distance difference based on the product of the first difference and the second difference.

[0134] In one embodiment of the present invention, the first wheel is the left front wheel, and the second wheel is the right front wheel; or, the first wheel is the left rear wheel, and the second wheel is the right rear wheel.

[0135] In one embodiment of the present invention, the second processing module 13 constructs a parking map according to the track information and the parking space information, including: correcting the track error of the track information according to the parking space information to obtain corrected target track information; and constructing a parking map according to the target track information.

[0136] In one embodiment of the present invention, the second processing module 13 corrects the track error of the track information according to the parking space information to obtain the corrected target track information, including: obtaining a first key frame, wherein when the parking space information of a new available parking space is detected based on the bird's-eye view, it is determined that the new available parking space is detected, and a first key frame is formed, and the first key frame includes the parking space information of the new available parking space and the newly added drivable area; constructing a second key frame according to the proportion of the new available parking spaces appearing in the first key frame; determining the correspondence between the first key frame and the second key frame, wherein the correspondence includes the intersection-and-union ratio between the parking spaces in the first key frame and the parking spaces in the second key frame; according to the correspondence, correcting the track error of the track information by a preset position correction algorithm to obtain the corrected target track information, wherein the track error includes the track motion error generated during the vehicle movement and the track calculation error generated during the track calculation.

[0137] In one embodiment of the present invention, the second processing module 13 constructs a parking space map according to the target track information, including: fusing the target track information with the parking space information in the second key frame to form a parking space map.

[0138] In one embodiment of the present invention, after forming the parking map, the second processing module 13 is further used to correct errors in the parking information of the newly added available parking spaces in the parking map by using a preset filtering algorithm to obtain a corrected parking map.

[0139] In one embodiment of the present invention, the second processing module 13 constructs a drivable area map according to the track information and the drivable area, including: based on a preset map fusion algorithm, fusing the target track information and the drivable area to obtain a drivable area map.

[0140] In one embodiment of the present invention, the second processing module 13 fuses the target track information and the drivable area based on a preset map fusion algorithm, including: obtaining obstacle information detected at multiple different times during the movement of the vehicle; based on a preset map fusion algorithm, fusing the obstacle information detected at multiple different times, and restoring the real position and size of the obstacle on a two-dimensional grid map to obtain a drivable area map.

[0141] In one embodiment of the present invention, the preset map fusion algorithm includes a map fusion algorithm implemented based on a Bayesian function.

[0142] In one embodiment of the present invention, the second processing module 13 obtains the parking map according to the parking space map and the drivable area map, including: merging the parking space map and the drivable area map to obtain the parking map.

[0143] In one embodiment of the present invention, the acquisition module 11 acquires the bird's-eye view of the vehicle, including: acquiring multiple initial pictures around the vehicle; and splicing the multiple initial pictures to obtain the bird's-eye view.

[0144] In one embodiment of the present invention, the acquisition module 11 stitches multiple initial images to obtain a bird's-eye view, including: based on an inverse perspective transformation algorithm, converting the multiple initial images into a first coordinate system, wherein the first coordinate system uses the ground as a coordinate plane and the center of the rear axle of the vehicle as a coordinate origin; projecting each point in the first coordinate system into a preset bird's-eye view template to obtain a bird's-eye view.

[0145] In one embodiment of the present invention, the acquisition module 11 acquires a plurality of initial images around the vehicle, including: acquiring a plurality of initial images around the vehicle in a one-to-one correspondence through a plurality of cameras distributed around the vehicle.

[0146] In one embodiment of the present invention, the device 1 for constructing a parking map further includes a control module (not shown in the figure). The control module is used to receive an external angle adjustment instruction; in response to the angle adjustment instruction, adjust the angle between the lens of at least one of the multiple cameras and the ground.

[0147] In one embodiment of the present invention, the initial picture comprises a wide-angle picture.

[0148] According to the device 1 for constructing a parking map of the embodiment of the present invention, the acquisition module 11 first acquires a bird's-eye view of the vehicle, and the bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle. The first processing module 12 can then determine the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle according to the bird's-eye view. The second processing module 13 constructs a parking map according to the parking space information and the drivable area, thereby facilitating automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0149] A further embodiment of the present invention also discloses a vehicle 2 .

[0150] In one embodiment of the present invention, Figure 6 As shown, the vehicle 2 includes the device 1 for constructing a parking map as described in the above-mentioned second aspect embodiment of the present invention.

[0151] In another embodiment of the present invention, the vehicle 2 includes: a processor, a memory, and a program for constructing a parking map stored in the memory and executable on the processor. When the program for constructing a parking map is executed by the processor, the method for constructing a parking map described in any one of the above embodiments of the present invention is implemented.

[0152] According to the vehicle 2 of the embodiment of the present invention, a bird's-eye view of the vehicle is first obtained. The bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle, and then the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle can be determined according to the bird's-eye view. A parking map is constructed according to the parking space information and the drivable area, so as to facilitate automatic parking. Therefore, the present invention only needs to identify the parking space information through the bird's-eye view, so the number of features required to be detected and identified is small, so the algorithm is simple, the amount of calculation is small, and the required computing power resources are small; on the other hand, due to the high stability of the parking space information, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, so the parking space information is easier to accurately identify and match, and the interference information can be effectively removed, so the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced, so the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0153] A further embodiment of the present invention also discloses a computer-readable storage medium.

[0154] The computer-readable storage medium of the embodiment of the present invention stores a program for constructing a parking map. When the program for constructing a parking map is executed by a processor, the method for constructing a parking map described in any one of the above embodiments of the present invention is implemented.

[0155] According to the computer-readable storage medium of the embodiment of the present invention, when the program for constructing a parking map stored thereon is executed by the processor, the bird's-eye view of the vehicle is first obtained, and the bird's-eye view of the vehicle provides environmental information within a preset range around the vehicle, and then the parking space information of the available parking spaces around the vehicle and the drivable area of ​​the vehicle can be determined according to the bird's-eye view. At the same time, the track information of the vehicle is obtained, and the parking space information and the track information are combined to construct a parking space map. Then, the track information is combined with the drivable area to construct a drivable area map, which can show the driver the area where the vehicle can drive safely and without obstacles. Then, the parking space map and the drivable area map are combined to obtain a comprehensive parking map, so that the driver can more accurately locate the available parking space according to the parking map, select the appropriate drivable area, and improve the parking accuracy of the vehicle. In addition, in constructing the parking map, there is no need to perform a large number of calculations, and the vehicle does not need to provide high computing power support, thereby reducing the amount of calculation. The parking map is constructed according to the parking space information and the drivable area, so as to facilitate automatic parking. Therefore, the present invention only needs to identify the parking space information through a bird's-eye view, so the number of features required to be detected and identified is relatively small. Therefore, the algorithm is simple, the amount of calculation is small, and the required computing power resources are small. On the other hand, since the parking space information is relatively stable, for example, the size, shape and other features of adjacent parking spaces are basically the same, and the error is very small, the parking space information is easier to accurately identify and match, and interference information can be effectively removed. Therefore, the stability and accuracy are better during detection and matching, and the probability of false detection and false matching can be effectively reduced. Therefore, the accuracy is higher when locating the parking space, and it is not easy to cause positioning loss.

[0156] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example.

[0157] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for constructing a parking map, characterized in that: include: Acquire a bird's-eye view of the vehicle, wherein the bird's-eye view includes environmental information within a preset range around the vehicle; Determining parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle according to the bird's-eye view; Obtaining a parking map according to the parking space information and the drivable area; Wherein, obtaining a parking map according to the parking space information and the drivable area includes: Acquiring the track information of the vehicle; the acquiring the track information of the vehicle comprises: acquiring the speed and heading angle of the vehicle; obtaining the track information according to the speed and heading angle; Constructing a parking space map according to the track information and the parking space information; constructing a parking space map according to the track information and the parking space information, including: correcting the track error of the track information according to the parking space information to obtain corrected target track information; constructing a parking space map according to the target track information; wherein, correcting the track error of the track information according to the parking space information to obtain corrected target track information includes: obtaining a first key frame, wherein, when parking space information of a new available parking space is detected based on the bird's-eye view, it is determined that the new available parking space is detected, and the first key frame is formed, and the first key frame Including parking space information of the new available parking space and the newly added drivable area; constructing a second key frame according to the proportion of the new available parking space appearing in the first key frame; determining the correspondence between the first key frame and the second key frame, wherein the correspondence includes the intersection-and-union ratio between the parking space in the first key frame and the parking space in the second key frame; according to the correspondence, correcting the track error of the track information by a preset position correction algorithm to obtain the corrected target track information, wherein the track error includes the track motion error generated during the movement of the vehicle and the track calculation error generated during the track calculation; Constructing a drivable area map according to the track information and the drivable area; The parking map is obtained according to the parking space map and the drivable area map.

2. The method for constructing a parking map according to claim 1, characterized in that: The determining, based on the bird's-eye view, parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle includes: The bird's-eye view is used as the input of a trained multi-task deep learning model to obtain the parking space information and the drivable area output by the multi-task deep learning model.

3. The method for constructing a parking map according to claim 2, characterized in that: The multi-task deep learning model includes a parking space information detection sub-model and a drivable area detection sub-model. The bird's-eye view is used as an input of the trained multi-task deep learning model to obtain the parking space information and the drivable area output by the multi-task deep learning model, including: Using the bird's-eye view as input to the parking space information detection sub-model, obtaining the parking space information output by the parking space information detection sub-model, wherein the parking space information includes parking space corner points; The bird's-eye view image is used as an input of the drivable area detection sub-model to obtain the drivable area output by the drivable area detection sub-model.

4. The method for constructing a parking map according to claim 2, characterized in that: Before using the bird's eye view image as input to the trained multi-task deep learning model, it also includes: When it is determined that the bird's-eye view image satisfies a preset output condition, using the bird's-eye view image as an input of a trained multi-task deep learning model; When it is determined that the bird's-eye view does not meet the preset output condition, the bird's-eye view is acquired again.

5. The method for constructing a parking map according to claim 4, characterized in that: Determining whether the bird's-eye view image meets a preset output condition includes: When it is determined that the set parameter for characterizing the image quality in the bird's-eye view meets the preset parameter threshold, it is determined that the bird's-eye view meets the preset output condition.

6. The method for constructing a parking map according to claim 1, characterized in that: The step of constructing a parking space map according to the target track information includes: The target track information is merged with the parking space information in the second key frame to form the parking space map.

7. The method for constructing a parking map according to claim 6, characterized in that: After forming the parking space map, the method further includes: The errors of the parking space information of the newly added available parking spaces in the parking space map are corrected by a preset filtering algorithm to obtain a corrected parking space map.

8. The method for constructing a parking map according to claim 1, characterized in that: The step of constructing a drivable area map according to the track information and the drivable area includes: Based on a preset map fusion algorithm, the target track information and the drivable area are fused to obtain a map of the drivable area.

9. The method for constructing a parking map according to claim 8, characterized in that: The method of fusing the target track information and the drivable area based on a preset map fusion algorithm includes: Obtain obstacle information detected at multiple different times during vehicle movement; Based on the preset map fusion algorithm, the obstacle information detected at multiple different times is fused, and the real position and size of the obstacle are restored on the two-dimensional grid map to obtain the drivable area map.

10. The method for constructing a parking map according to claim 1, characterized in that: The obtaining of the parking map according to the parking space map and the drivable area map includes: The parking space map and the drivable area map are combined to obtain the parking map.

11. The method for constructing a parking map according to any one of claims 1, characterized in that: The obtaining of a bird's-eye view of the vehicle comprises: Acquire a plurality of initial images around the vehicle; Multiple initial images are stitched together to obtain the bird's-eye view.

12. A device for constructing a parking map, characterized in that: include: An acquisition module, used to acquire a bird's-eye view of the vehicle, wherein the bird's-eye view includes environmental information within a preset range around the vehicle; A first processing module, configured to determine parking space information of available parking spaces around the vehicle and a drivable area of ​​the vehicle according to the bird's-eye view; A second processing module, configured to obtain a parking map according to the parking space information and the drivable area; Wherein, obtaining a parking map according to the parking space information and the drivable area includes: Acquiring the track information of the vehicle; the acquiring the track information of the vehicle comprises: acquiring the speed and heading angle of the vehicle; obtaining the track information according to the speed and heading angle; Constructing a parking space map according to the track information and the parking space information; constructing a parking space map according to the track information and the parking space information, including: correcting the track error of the track information according to the parking space information to obtain corrected target track information; constructing a parking space map according to the target track information; wherein, correcting the track error of the track information according to the parking space information to obtain corrected target track information includes: obtaining a first key frame, wherein, when parking space information of a new available parking space is detected based on the bird's-eye view, it is determined that the new available parking space is detected, and the first key frame is formed, and the first key frame Including parking space information of the new available parking space and the newly added drivable area; constructing a second key frame according to the proportion of the new available parking space appearing in the first key frame; determining the correspondence between the first key frame and the second key frame, wherein the correspondence includes the intersection-and-union ratio between the parking space in the first key frame and the parking space in the second key frame; according to the correspondence, correcting the track error of the track information by a preset position correction algorithm to obtain the corrected target track information, wherein the track error includes the track motion error generated during the movement of the vehicle and the track calculation error generated during the track calculation; Constructing a drivable area map according to the track information and the drivable area; The parking map is obtained according to the parking space map and the drivable area map.

13. A vehicle, characterized in that: include: The device for constructing a parking map as claimed in claim 12; or, A processor, a memory, and a program for constructing a parking map stored in the memory and executable on the processor, wherein the program for constructing a parking map implements the method for constructing a parking map according to any one of claims 1 to 11 when executed by the processor.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program for constructing a parking map, and when the program for constructing a parking map is executed by a processor, the method for constructing a parking map according to any one of claims 1 to 11 is implemented.

Images