Parking space identification method, device, computer equipment and readable storage medium

Abstract

The present application relates to a parking space identification method, device, computer equipment and readable storage medium; obtaining a storage space feature map and a key point heat map of a storage space image to be processed; predicting the storage space feature map to obtain an initial storage space identification result; determining a target key point that matches the first position information of an entry corner point from the key point heat map, and determining the position information of the target key point as the second position information of the entry corner point; the accuracy of the second position information is greater than the first position information; determining the target parking space according to the second position information of the entry corner point, the storage space direction and the storage space frame information. Based on the key point heat map and the first position information of the entry corner point, the second position information corresponding to the entry corner point can be accurately matched, thereby enhancing the accuracy and reliability of the entry corner point positioning, and then combining the storage space direction and the storage space frame information to accurately determine the target parking space, effectively improving the accuracy and reliability of storage space identification.

Description

Technical Field

[0001] The present application relates to the field of intelligent parking technology, and in particular to a parking space identification method, device, computer equipment and readable storage medium. Background Art

[0002] With the rapid development of intelligent driving, automatic parking has become one of the indispensable functions. Among them, parking space recognition is one of the core technologies, which aims to search for parking spaces within the visible range. In the process of automatic parking, the existing technology needs to identify the positions and types of the four corner points of the parking space, and constrain the positions of the four corner points according to geometric features to realize the planning of the vehicle's automatic parking path.

[0003] However, the existing method of using four corner points to identify the storage location has the problem of poor accuracy of the storage location identification result. Summary of the invention

[0004] Based on this, it is necessary to provide a parking space identification method, device, computer equipment and readable storage medium to address the above technical problems.

[0005] In a first aspect, the present application provides a parking space identification method, the method comprising:

[0006] Obtain the storage location feature map and key point heat map of the storage location image to be processed;

[0007] Predicting the storage location feature map to obtain an initial storage location recognition result; wherein the initial storage location recognition result includes first position information of an entrance corner point corresponding to the storage location, storage location direction, and storage location frame information;

[0008] Determine a target key point that matches the first position information of the entry corner point from the key point heat map, and determine the position information of the target key point as the second position information of the entry corner point; the accuracy of the second position information is greater than that of the first position information;

[0009] A target parking space is determined according to the second position information of the entrance corner point, the storage location direction and the storage location frame information.

[0010] In one embodiment, the step of obtaining a storage location feature map and a key point heat map of the storage location image to be processed includes:

[0011] Get the image of the storage location to be processed;

[0012] The storage location image to be processed is input into a trained parking space detection model, and feature extraction is performed through a feature extraction network in the trained parking space detection model to generate a storage location feature map and a key point heat map; wherein the feature extraction network includes a self-calibration convolution layer.

[0013] In one embodiment, the storage location feature map includes storage location feature information; and predicting the storage location feature map to obtain an initial storage location identification result includes:

[0014] The storage location feature map is input into the prediction network in the trained parking space detection model, and the storage location feature information is predicted by the prediction network to obtain an initial storage location recognition result.

[0015] In one embodiment, the method further comprises:

[0016] Obtain multiple annotated sample parking space bird's-eye views and an initial parking space detection model;

[0017] Constructing a target loss function according to the labeled sample storage location bird's-eye view;

[0018] According to the labeled sample parking space bird's-eye view, the initial parking space detection model is trained until the target loss function meets a preset convergence condition, thereby obtaining a trained parking space detection model;

[0019] The constructing a target loss function according to the labeled sample storage location bird's-eye view includes:

[0020] Inputting the labeled sample storage location bird's-eye view into the initial parking space detection model to obtain a sample storage location recognition result;

[0021] According to the sample storage location identification results and the data labels corresponding to the labeled sample storage location bird's-eye view, a storage location entry corner point loss function, a storage location direction loss function, a storage location frame loss function, a storage location depth loss function, a storage location category loss function and a limiter loss function are respectively constructed;

[0022] The target loss function is obtained by performing a weighted sum calculation on the storage location entrance corner point loss function, the storage location direction loss function, the storage location frame loss function, the storage location depth loss function, the storage location category loss function and the limiter loss function.

[0023] In one embodiment, the data label includes a heat map label; the sample storage location identification result includes a storage location entrance corner point prediction result; and constructing a storage location entrance corner point loss function includes:

[0024] The storage location entrance corner point loss function is determined according to the heat map label and the prediction point confidence in the storage location entrance corner point prediction result.

[0025] In one embodiment, the data label includes a storage location direction label; the sample storage location identification result includes a storage location direction prediction result; and constructing a storage location direction loss function includes:

[0026] Determine a confidence loss function according to the approximate direction angle true value and the angle offset true value in the storage location direction label, and the approximate direction angle prediction value and the angle offset prediction value in the storage location direction prediction result;

[0027] Determine a positioning loss function according to the true value of the angle offset in the storage location direction label and the predicted value of the angle offset in the storage location direction prediction result;

[0028] The confidence loss function and the positioning loss function are weightedly summed to obtain a storage location direction loss function.

[0029] In one of the embodiments, determining the target key point matching the first position information of the entry corner point from the key point heat map includes:

[0030] Based on a preset position mapping relationship and the first position information of the entry corner point, a target key point matching the first position information is determined from the key point heat map.

[0031] In a second aspect, the present application further provides a parking space identification device, the device comprising:

[0032] An acquisition module is used to acquire a storage location feature map and a key point heat map of the storage location image to be processed;

[0033] An identification module is used to predict the storage location feature map to obtain an initial storage location identification result; wherein the initial storage location identification result includes the first position information of the entrance corner point corresponding to the storage location, the storage location direction and the storage location frame information;

[0034] a position information correction module, used to determine a target key point matching the first position information of the entry corner point from the key point heat map, and determine the position information of the target key point as the second position information of the entry corner point; the accuracy of the second position information is greater than that of the first position information;

[0035] The storage location determination module is used to determine the target parking space according to the second position information of the entrance corner point, the storage location direction and the storage location frame information.

[0036] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in any one of the embodiments of the first aspect are implemented.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in any one of the embodiments of the first aspect above are implemented.

[0038] In a fifth aspect, the present application further provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the method described in any one of the embodiments of the first aspect above.

[0039] The above-mentioned parking space identification method, device, computer equipment and readable storage medium, based on the storage space feature map of the storage space image to be processed, realizes preliminary prediction of the storage space image to be processed from a global perspective, based on which the potential storage space area can be quickly located; further, based on the key point heat map corresponding to the storage space image to be processed and the first position information of the entrance corner point, the second position information corresponding to the entrance corner point can be accurately matched, and the accuracy of the second position information is greater than the first position information, thereby realizing high-precision positioning of the entrance corner point, enhancing the accuracy and reliability of the entrance corner point positioning, and laying a foundation for improving the accuracy of the target parking space identification result; further, based on the second position information corresponding to the entrance corner point, the storage space direction and the storage space frame information, the target parking space can be accurately determined, and the position information of the storage space tail corner point corresponding to the target parking space can be accurately calculated, thereby avoiding the problem in the prior art that the storage space is identified by relying on four corner points, and the target parking space cannot be accurately identified when the storage space tail corner point is blocked or blurred, thereby effectively improving the accuracy and reliability of storage space identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A diagram showing an application environment of a parking space identification method in an embodiment;

[0042] Figure 2 A schematic diagram of a process of a parking space identification method in one embodiment;

[0043] Figure 3 A schematic diagram of a vehicle's surrounding environment view in one embodiment;

[0044] Figure 4 A schematic diagram of a storage location image to be processed in one embodiment;

[0045] Figure 5 A schematic diagram of a parking space in one embodiment;

[0046] Figure 6 is a schematic diagram of different types of parking spaces in one embodiment;

[0047] Figure 7 A schematic diagram of a three-dimensional structure of a Gaussian distribution and a Laplace distribution in one embodiment;

[0048] Figure 8 Schematic diagram of the training process of the initial parking space detection model in one embodiment;

[0049] Fig. 9 A schematic diagram of a multi-box classification method in one embodiment;

[0050] Fig.10 A schematic diagram of a flow chart of a parking space identification method in a specific embodiment;

[0051] Fig.11 is a structural block diagram of a parking space identification device in one embodiment;

[0052] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] The parking space identification method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, tablet computers and Internet of Things devices, and the Internet of Things devices can be smart car-mounted devices, projection devices, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0055] In an exemplary embodiment, Figure 2 As shown, a parking space identification method is provided. This embodiment takes the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0056] Step S201, obtaining a storage location feature map and a key point heat map of the storage location image to be processed.

[0057] The image of the storage space to be processed refers to a bird's-eye view corresponding to the parking space; it should be noted that the parking space is usually presented as a regular quadrilateral in the bird's-eye view.

[0058] Exemplarily, multiple camera devices are arranged on the vehicle body to collect multiple views of the vehicle's surrounding environment; wherein the camera devices may include but are not limited to cameras; the views of the vehicle's surrounding environment are fisheye images, such as Figure 3 As shown; further, multiple views of the vehicle's surrounding environment are spliced ​​to obtain a parking space image to be processed, that is, a bird's-eye view corresponding to the parking space, such as Figure 4 Optionally, the method of splicing multiple vehicle surrounding environment views to obtain the storage location image to be processed may be, but is not limited to: performing inverse perspective transformation on multiple vehicle surrounding environment views to convert the multiple vehicle surrounding environment views into the storage location image to be processed.

[0059] Among them, the warehouse location feature map refers to the feature map obtained by extracting features from the warehouse location image to be processed; the warehouse location feature map includes warehouse location feature information in different scale spaces; the key point heat map is a two-dimensional matrix, and each pixel value in the key point heat map is used to characterize the possibility that the pixel point is a key point; wherein the pixel value is the response intensity or confidence, and generally the higher the pixel value, the more likely the location is a key point.

[0060] Step S202, predicting the storage location feature map to obtain an initial storage location recognition result.

[0061] The initial storage location recognition result includes the first position information of the entrance corner point corresponding to the storage location, the storage location direction and the storage location frame information.

[0062] Among them, the entrance corner points refer to the two corner points corresponding to the entrance direction of the parking space; the first position information of the entrance corner points refers to the position information obtained by preliminary prediction of the entrance corner points based on the storage space feature map; wherein the position information refers to the coordinate position of the entrance corner points in the storage space image to be processed. The storage space direction refers to the direction pointing to the inside of the parking space along the entrance corner points with the entrance corner points as the origin; it can be understood that the storage space direction can be used to determine the angle at which the vehicle is facing when entering the storage space. The storage space frame information includes the coordinates of the center point of the storage space frame, the width of the storage space frame, and the height of the storage space frame; wherein the storage space frame width refers to the width of the circumscribed rectangle of the storage space frame; the storage space frame height is equal to the height of the circumscribed rectangle of the storage space frame.

[0063] For example, Figure 5 Take the following as an example, where the storage location direction is The width of the storage location frame is equal to the width of the circumscribed rectangle 100 of the storage location frame, and the height of the storage location is equal to the height h of the circumscribed rectangle 100 of the storage location frame.

[0064] It should be noted that the initial storage location identification result may also include storage location depth, storage location category, and limiter position information, etc., which need to be identified according to actual needs and are not specifically limited here.

[0065] The depth information of the storage location is equal to the distance between the corner point of the same side entrance and the corner point of the tail end; the tail end corner point is the corner point corresponding to the bottom of the storage location. The storage location category can include but is not limited to vertical storage locations, lateral storage locations, and oblique storage locations; for example, Figure 6 As shown, storage location 1 is a vertical storage location, storage location 2 is an inclined storage location, and storage location 3 is a lateral storage location, where d represents the depth of the storage location.

[0066] Step S203: determine a target key point that matches the first position information of the entry corner point from the key point heat map, and determine the position information of the target key point as the second position information of the entry corner point.

[0067] The accuracy of the second position information is greater than that of the first position information.

[0068] Among them, the target key point refers to the key point with the highest correlation with the entry corner point in the key point heat map.

[0069] Among them, the key point heat map has a high spatial resolution and can fully retain local spatial information, so it can achieve high-precision positioning of the entry corner point; it should be noted that although the key point heat map can accurately derive the possibility that each pixel point is a key point (that is, the higher the response intensity or confidence corresponding to the pixel point, the higher the possibility that the position is a key point), the key point heat map may contain key points corresponding to non-entry corner points. In contrast, the storage location feature map is based on a global perspective and helps to predict and identify potential entry corner point locations. Therefore, by combining the preliminary positioning information provided by the storage location feature map (that is, the first position information of the entry corner point) and the key point heat map, it is possible to accurately match the high response intensity target key point that is most relevant to the entry corner point, and determine the position information of the target key point as the second position information of the entry corner point, so as to enhance the accuracy of the entry corner point positioning.

[0070] Step S204: determining the target parking space according to the second position information of the entrance corner point, the storage location direction and the storage location frame information.

[0071] Exemplarily, there may be multiple parking spaces in the storage space image to be processed. Therefore, when determining the target parking space, it is necessary to associate the entrance corner point corresponding to the target parking space with the corresponding storage space frame information, and then, based on the entrance corner point corresponding to the target parking space, the storage space direction and the storage space frame information, calculate the position information of the storage space tail corner point corresponding to the target parking space, based on which the target parking space can be accurately determined.

[0072] It should be noted that the method for associating the entrance corner point corresponding to the target parking space with the corresponding storage space frame information may be, but is not limited to, using cluster analysis to associate the entrance corner point corresponding to the target parking space with the corresponding storage space frame information, which is not specifically limited here.

[0073] In the present embodiment, based on the storage location feature map of the storage location image to be processed, a preliminary prediction of the storage location image to be processed is achieved from a global perspective, based on which the potential storage location area can be quickly located; further, based on the key point heat map corresponding to the storage location image to be processed and the first position information of the entrance corner point, the second position information corresponding to the entrance corner point can be accurately matched, and the accuracy of the second position information is greater than the first position information, thereby achieving high-precision positioning of the entrance corner point, enhancing the accuracy and reliability of the entrance corner point positioning, and laying a foundation for improving the accuracy of the target parking space identification result; further, based on the second position information corresponding to the entrance corner point, the storage location direction and the storage location frame information, the target parking space can be accurately determined, and the position information of the storage location tail corner point corresponding to the target parking space can be accurately calculated, thereby avoiding the problem in the prior art of relying on four corner points to identify the storage location, and failing to accurately identify the target parking space when the storage location tail corner point is blocked or blurred, thereby effectively improving the accuracy and reliability of storage location identification.

[0074] In one embodiment, obtaining a storage location feature map and a key point heat map of a storage location image to be processed includes the following steps:

[0075] Step 1: Get the image of the storage location to be processed.

[0076] The storage space image to be processed refers to the bird's-eye view corresponding to the parking space.

[0077] Optionally, the method for obtaining the image of the storage location to be processed may be: obtaining a plurality of views of the vehicle's surrounding environment; and performing splicing processing on the plurality of views of the vehicle's surrounding environment to obtain the image of the storage location to be processed.

[0078] Step 2: Input the image of the storage location to be processed into the trained parking space detection model, perform feature extraction through the feature extraction network in the trained parking space detection model, and generate a storage location feature map and a key point heat map.

[0079] Among them, the trained parking space detection model includes a feature extraction network and a prediction network; wherein, the feature extraction network includes a self-calibration convolution layer; it should be noted that the number of self-calibration convolution layers needs to be set according to the actual needs of the model, and no specific limitation is made here.

[0080] The self-calibrated convolution layer allows homogeneous and independent feature transformation processes; specifically, based on the self-calibrated convolution layer, convolutional feature transformation can be performed in different scale spaces, so that the feature map and the input share the original scale space and the downsampled hidden space with the same resolution. Therefore, by adding a self-calibrated convolution layer to the feature extraction network, it is possible to adaptively establish long-range and inter-channel dependencies around each spatial position during the feature extraction process, so as to explicitly merge rich information to generate discriminative representations, thereby making the feature extraction network more representative.

[0081] Specifically, an image of the storage location to be processed is obtained, and the image of the storage location to be processed is input into a trained parking space detection model. After feature extraction through the feature extraction network in the trained parking space detection model, the image is divided into two branches, one for generating a storage location feature map for subsequent prediction, and the other for generating a key point heat map.

[0082] Further, in terms of generating a key point heat map, it is preferred to use Labrador distribution instead of Gaussian distribution in the prior art to construct a key point heat map; Figure 7 As shown, Figure 7 The left side of the figure shows the three-dimensional structure of Gaussian distribution, and the right side shows the three-dimensional structure of Labrador distribution. Figure 7 It can be seen that at the same key point, the key point response based on Laplace distribution is more concentrated, based on which the accuracy of entrance corner point positioning can be further improved.

[0083] In this embodiment, based on the self-calibrated convolutional layer, in the process of feature extraction of the storage location image to be processed, the long-range and channel dependencies can be adaptively established around each spatial position, so as to explicitly merge rich information to generate a discriminative representation, thereby making the feature extraction network more representative; based on this, the storage location image to be processed is input into the trained parking space detection model, and feature extraction is performed through the feature extraction network in the trained parking space detection model, so that a more reliable storage location feature map and key point heat map can be generated, which lays a foundation for improving the accuracy of parking space recognition.

[0084] In one embodiment, the storage location feature map includes storage location feature information; wherein the storage location feature information includes low-level feature information and high-level feature information.

[0085] Predict the location feature map and obtain the initial location recognition results, including:

[0086] The storage location feature map is input into the prediction network in the trained parking space detection model, and the storage location feature information is predicted by the prediction network to obtain the initial storage location recognition result.

[0087] Among them, the prediction network is used to perform various types of prediction tasks.

[0088] Exemplarily, the image of the storage location to be processed is input into a trained parking space detection model, and feature extraction is performed through the feature extraction network in the trained parking space detection model to generate a storage location feature map. Further, the storage location feature map is input into the prediction network in the trained parking space detection model, and the storage location feature information is predicted through the prediction network to obtain an initial storage location recognition result.

[0089] In this embodiment, based on the parking space feature map and the prediction network, an accurate initial parking space identification result can be generated, which lays a foundation for accurately determining the target parking space.

[0090] In one embodiment, Figure 8 As shown, Figure 8 FIG. 1 is a schematic diagram of a training process of an initial parking space detection model in an embodiment; the parking space recognition method further includes the following steps:

[0091] Step S801, obtaining a plurality of labeled sample parking space bird's-eye views and an initial parking space detection model.

[0092] The initial parking space detection model includes a feature extraction network and a prediction network. Optionally, the initial parking space detection model can be constructed based on a neural network framework, and a self-calibration convolution layer is added to the feature extraction network to enhance the model's ability to represent entrance corner features and other information.

[0093] In an exemplary embodiment, before obtaining multiple annotated sample storage location bird's-eye views, the method also includes: obtaining multiple groups of sample vehicle surrounding environment views; preprocessing the multiple groups of sample vehicle surrounding environment views to obtain initial storage location bird's-eye views corresponding to each group of sample vehicle surrounding environment views; screening the multiple initial storage location bird's-eye views according to preset conditions to obtain multiple sample storage location bird's-eye views; wherein the sample storage location bird's-eye view includes multiple parking spaces; it should be noted that the preprocessing of the multiple groups of sample vehicle surrounding environment views may include but is not limited to denoising and inverse perspective transformation, which are not specifically limited here; wherein the preset conditions may include but are not limited to using the initial storage location bird's-eye view including multiple parking spaces as the sample storage location bird's-eye view.

[0094] Among them, the labeled sample storage location bird's-eye view refers to the image after the sample storage location bird's-eye view is labeled; it can be understood that there are corresponding data labels for the labeled sample storage location bird's-eye view, wherein the data labels may include but are not limited to entrance corner position labels, storage location direction labels, storage location frame labels, storage location depth labels, storage location category labels, limiter position labels, and heat map labels.

[0095] In an exemplary embodiment, when determining the heat map label corresponding to the sample storage location bird's-eye view map, the sample heat map corresponding to the sample storage location bird's-eye view map is recorded as ;in, Indicates the width of the bird's-eye view of the sample storage location. Indicates the height of the bird's-eye view of the sample storage location, Indicates the scaling factor of the sample heat map relative to the sample location bird's-eye view. Furthermore, the entrance corner points in the sample location bird's-eye view are regarded as key points. , for each key point , which can calculate key points The corresponding low-resolution equivalent on the sample heat map ,in , further, based on the Laplace distribution, a heat map that obeys the Laplace distribution is generated for each key point; it can be understood that by superimposing the heat maps corresponding to all key points, the sample heat map corresponding to the bird's-eye view of the sample storage location can be obtained, and then the corresponding heat map label can be determined based on the sample heat map; wherein, the heat map label is the response intensity or confidence of each pixel in the sample heat map, and the heat map label can be calculated by formula (1).

[0096] (1)

[0097] in, , represents the response intensity or confidence level corresponding to each pixel in the sample heat map, , represents the key point adaptive scale parameter, which is used to control the diffusion degree of the response intensity peak; Represents the coordinates of the predicted point in the predicted heat map corresponding to the sample heat map; It indicates the coordinates of the low-resolution equivalent key points. It should be noted that whether the predicted point in the predicted heat map is an entry corner point needs to be determined based on the response intensity or confidence corresponding to the predicted point.

[0098] In an exemplary embodiment, based on a multi-box classification method, the location direction regression problem is transformed into a location direction classification problem, which can avoid the problem of poor accuracy in location direction recognition caused by direct direction regression in the prior art and further improve the accuracy of location direction recognition.

[0099] Among them, the multi-box classification method aims to Divide into Bins are combined with the direction angle deviation value to complete the expression of the direction. That is, when labeling the storage location direction corresponding to the bird's-eye view of the sample storage location, the approximate direction angle true value and the angle offset true value are used to represent the storage location direction; for example, Fig. 9 As shown, based on the multi-box classification method, Divided into 8 bins; storage direction , which can be expressed as ;in, is the approximate true value of the direction angle, is the true value of the angle relative to the approximate direction The true value of the angle offset; it should be noted that the storage location direction is the storage location direction corresponding to the i-th entrance corner point; approximate direction angle true value Determined by the divided bins; optional, approximate direction angle true value , which can be the center angle of the corresponding bins, for example, the range is The approximate direction angle corresponding to the box Bins Can be .

[0100] Step S802: construct a target loss function based on the labeled bird's-eye view of sample storage locations.

[0101] It is understandable that in multi-task learning, the target loss function is used to optimize multiple related tasks at the same time; each task has a corresponding loss function, and the target loss function is determined by the weighted sum of the loss functions corresponding to multiple tasks. Based on the target loss function, it can ensure that the model can perform well on all tasks.

[0102] Step S803 , training the initial parking space detection model according to the labeled sample parking space bird's-eye view until the target loss function meets the preset convergence condition, thereby obtaining a trained parking space detection model.

[0103] Among them, the preset convergence conditions need to be set according to the model training requirements and are not specifically limited here.

[0104] Specifically, multiple groups of sample vehicle surrounding environment views are obtained; the multiple groups of sample vehicle surrounding environment views are preprocessed to obtain initial storage location bird's-eye view corresponding to each group of sample vehicle surrounding environment views; according to preset conditions, multiple initial storage location bird's-eye view views are screened to obtain multiple sample storage location bird's-eye view views; further, the multiple sample storage location bird's-eye view views are labeled to obtain multiple labeled sample storage location bird's-eye view views; further, based on the labeled sample storage location bird's-eye view views, loss functions corresponding to multiple tasks are respectively constructed; weighted sum calculation is performed on the loss functions of multiple tasks to determine the target loss function; further, based on the labeled sample storage location bird's-eye view views, the initial parking space detection model is trained until the target loss function meets the preset convergence conditions to obtain a trained parking space detection model.

[0105] In this embodiment, based on the labeled sample storage location bird's-eye view and the target loss function, the initial parking space detection model can ensure that the model can learn various characteristics of the storage location, thereby improving the generalization ability of the model. By ensuring that the target loss function meets the preset convergence condition during the training process, a trained parking space detection model is obtained, thereby laying the foundation for achieving high-precision and robust parking space recognition.

[0106] In one embodiment, constructing a target loss function based on the labeled sample storage location bird's-eye view includes the following steps:

[0107] Step 1: Input the labeled sample storage location bird's-eye view into the initial parking space detection model to obtain the sample storage location recognition result.

[0108] Step 2: Based on the sample storage location recognition results and the data labels corresponding to the annotated sample storage location bird's-eye view, the storage location entrance corner loss function, storage location direction loss function, storage location frame loss function, storage location depth loss function, storage location category loss function and limiter loss function are constructed respectively.

[0109] Step 3: Perform weighted sum calculation on the storage location entrance corner point loss function, storage location direction loss function, storage location frame loss function, storage location depth loss function, storage location category loss function and limiter loss function to obtain the target loss function.

[0110] For example, the storage location entrance corner loss function is recorded as , the location direction loss function is recorded as , the storage location frame loss function is recorded as , the storage location depth loss function is recorded as , storage location category loss function And the limiter loss function . Loss function for the corner point of the storage location entrance , storage location direction loss function , storage location box loss function , storage location depth loss function , storage location category loss function And the limiter loss function Perform weighted sum calculation to obtain the target loss function Loss, which is shown in formula (2).

[0111] (2)

[0112] Among them, a, b, c, d, e, and f are hyperparameters that control weights.

[0113] Further, preferably, the data label includes a heat map label; the sample storage location recognition result includes a storage location entrance corner point prediction result; and constructing a storage location entrance corner point loss function includes:

[0114] According to the heat map label and the confidence of the predicted point in the prediction result of the storage location entrance corner point, the storage location entrance corner point loss function is determined.

[0115] For example, the storage location entry corner loss function As shown in formula (3).

[0116] (3)

[0117] in, and Represents training hyperparameters, based on experience , ; N represents the number of key points in the bird's-eye view of the sample storage location after annotation, is the heat map label. For the specific calculation process, please refer to the above formula (1); Represents the true value label corresponding to the predicted point; If it is equal to 1, it means that the predicted point is the required key point, that is, the entry corner point; When it is equal to other values, it means that the current prediction point is not the required key point, that is, it is not the entry corner point.

[0118] Preferably, the data label includes a storage location direction label; the sample storage location identification result includes a storage location direction prediction result; and constructing a storage location direction loss function includes the following steps:

[0119] Step 1: Determine the confidence loss function according to the true value of the approximate direction angle and the true value of the angle offset in the storage location direction label, and the predicted value of the approximate direction angle and the predicted value of the angle offset in the storage location direction prediction result.

[0120] It should be noted that this embodiment is based on the multi-bin classification method, and uses the approximate direction angle true value and the angle offset true value to characterize the storage location direction. Therefore, the storage location direction label includes the approximate direction angle true value and the angle offset true value; wherein, the approximate direction angle true value can be the center angle of the corresponding box Bins; the angle offset true value refers to the angle offset relative to the approximate direction angle true value.

[0121] Step 2: Determine the positioning loss function according to the true value of the angle offset in the storage location direction label and the predicted value of the angle offset in the storage location direction prediction result.

[0122] Step 3: Perform weighted sum calculation on the confidence loss function and the positioning loss function to obtain the storage location direction loss function.

[0123] For example, the location direction loss function As shown in formula (4).

[0124] (4)

[0125] (5)

[0126] (6)

[0127] (7)

[0128] (8)

[0129] in, represents the confidence loss function; , represents the positioning loss function; It is a training hyperparameter, which needs to be determined based on experience; Indicates the storage location direction label corresponding to the i-th entry corner point; Represents the true value of the approximate direction angle; Represents the true value of the angle offset corresponding to the i-th entry corner point; It represents the prediction result of the storage location direction corresponding to the i-th entrance corner point; Represents the approximate direction angle prediction value; represents the predicted value of the angle offset corresponding to the i-th entry corner point; The divided boxes are Bins.

[0130] Preferably, the data label includes a storage location frame label; the sample storage location identification result includes a storage location frame prediction result; and a storage location frame loss function is constructed based on the storage location frame label and the storage location frame prediction result. , as shown in formula (9).

[0131] (9)

[0132] in, It represents the overlap ratio between the predicted value of the storage location box and the true value of the storage location box, that is, the ratio of the intersection to the union; Indicates the label of the storage location box, that is, the true value of the storage location box; Indicates the storage location box prediction result, that is, the storage location box prediction value.

[0133] Preferably, the data label also includes a storage location category label; the sample storage location identification result includes a storage location category prediction result; based on the storage location category label and the storage location category prediction result, a storage location category loss function is constructed. , as shown in formula (10).

[0134] (10)

[0135] in, Indicates the number of storage location categories; Represents the storage location category label, that is, the true value of the storage location category; Indicates the storage location category prediction result, that is, the predicted storage location category value.

[0136] Preferably, the data label also includes a location depth label; the sample location identification result includes a location depth prediction result; based on the location depth label and the location depth prediction result, a location depth loss function is constructed. , as shown in formula (11).

[0137] (11)

[0138] in, Indicates the number of storage locations whose depths need to be measured on the bird's-eye view of the labeled sample storage locations; The location depth label of each location is the true value of the location depth; Indicates the prediction result of the storage depth of each storage location, that is, the predicted depth value of each storage location; It represents a hyperparameter, and its empirical value is 0.5.

[0139] Preferably, the data label also includes a stopper position label; the sample storage location identification result includes a stopper position prediction result; and a stopper loss function is constructed based on the stopper position label and the stopper position prediction result. , as shown in formula (12).

[0140] (12)

[0141] in, The limiter position label represents the true value of the limiter position information; Indicates the stopper position prediction result, i.e. the predicted stopper position information.

[0142] In this embodiment, based on the sample storage location identification results and the data labels corresponding to the annotated sample storage location bird's-eye view, the storage location entrance corner point loss function, storage location direction loss function, storage location frame loss function, storage location depth loss function, storage location category loss function and limiter loss function can be constructed respectively. Based on this, the model's learning of each feature can be more finely optimized to ensure that the trained parking space detection model can more accurately and stably output the parking space location, orientation, depth, category, limiter and other information, thereby effectively improving the accuracy of parking space identification.

[0143] In one embodiment, determining a target key point that matches the first position information of the entry corner point from the key point heat map includes:

[0144] Based on the preset position mapping relationship and the first position information of the entry corner point, a target key point matching the first position information is determined from the key point heat map.

[0145] The preset position mapping relationship refers to the position mapping relationship between the image of the warehouse location to be processed, the warehouse location feature map corresponding to the image of the warehouse location to be processed, and the key point heat map. It should be noted that the warehouse location feature map and the key point heat map are both images obtained based on the warehouse location image to be processed. Therefore, for a specific location, there must be a mapping relationship between the three. In addition, it should be noted that the warehouse location image to be processed, the warehouse location feature map, and the key point heat map Figure 3 There may be a certain scaling relationship between them.

[0146] Exemplarily, when there is a scaling relationship between the image of the storage location to be processed and the key point heat map; based on the preset position mapping relationship and the first position information of the entry corner point, the target key point corresponding to the first position information of the entry corner point can be accurately matched from the key point heat map; then, based on the heat map coordinates of the target key point in the key point heat map, and combined with the scaling factor between the key point heat map and the image of the storage location to be processed, the position information of the target key point can be calculated, and then the position information of the target key point is used as the second position information of the entry corner point to improve the accuracy of the entry angle positioning.

[0147] In this embodiment, based on the first position information of the entry corner point, the potential entry corner point position can be located; then based on the preset position mapping relationship and the first position information of the entry corner point, the target key point corresponding to the first position information of the entry corner point can be quickly and accurately matched from the key point heat map, laying the foundation for improving the positioning accuracy of the entry corner point.

[0148] In a specific embodiment, see Fig.10 The parking space identification method comprises the following steps:

[0149] Step S1001, obtaining a plurality of vehicle surrounding environment views, and splicing the plurality of vehicle surrounding environment views to obtain a storage location image to be processed.

[0150] Step S1002: Input the image of the storage location to be processed into the feature extraction network in the trained parking space detection model to extract features, and generate a storage location feature map and a key point heat map corresponding to the image of the storage location to be processed.

[0151] Step S1003, inputting the storage location feature map into the prediction network in the trained parking space detection model, predicting the storage location feature information through the prediction network, and obtaining an initial storage location recognition result.

[0152] The initial location identification result includes the first position information and confidence level of the entrance corner point corresponding to the location, the location direction and confidence level, the location frame information and confidence level, the location depth information and confidence level, the location category and confidence level, and the position information and confidence level of the stopper. ,in Indicates the coordinates of the center point of the storage location frame. is the location frame width, is the height of the storage location frame; the storage location direction is ,in, Indicates the storage location direction corresponding to the first entry corner point. Indicates the storage location direction corresponding to the second entrance corner point; the storage location category is , the confidence level corresponding to the storage location category is ; The location depth is , the confidence level corresponding to the storage location depth is , the position information of the limiter is , the confidence level corresponding to the position information of the limiter is ;in, is the first limiter position information, is the confidence level of the first limiter position information; is the second limiter position information, is the confidence level of the second limit stopper position information.

[0153] Step S1004, based on the preset position mapping relationship and the first position information of the entry corner point, determine the target key point that matches the first position information of the entry corner point from the key point heat map; determine the position information of the target key point as the second position information of the entry corner point.

[0154] The second position information has a higher accuracy than the first position information. ,in, Second position information representing a first entry corner point; Indicates second position information of the second entry corner point.

[0155] Step S1005: determining a target parking space according to the second position information of the entrance corner point, the storage location direction and the storage location frame information.

[0156] Exemplarily, the second position information of the two entrance corner points corresponding to the target parking space and the storage location frame information are associated through clustering; then, based on the second position information of the entrance corner points, combined with the storage location direction and the storage location frame information, the position information of the tail corner point of the storage location is calculated; finally, based on the second position information of the entrance corner points, the storage location direction, the storage location frame information and the position information of the tail corner point, the target parking space can be accurately determined, thus avoiding the problem in the prior art that the target parking space cannot be identified when the tail corner point of the storage location is blocked or blurred.

[0157] It should be noted that in actual applications, by identifying the storage location depth, storage location category, and limiter position information, corresponding data support can be provided for downstream needs.

[0158] In the above-mentioned parking space identification method, on the one hand, based on the storage space feature map of the storage space image to be processed, a preliminary prediction of the storage space image to be processed is realized from a global perspective, based on which the potential storage space area can be quickly located; further, based on the key point heat map corresponding to the storage space image to be processed and the first position information of the entrance corner point, the second position information corresponding to the entrance corner point can be accurately matched, and the accuracy of the second position information is greater than the first position information, thereby realizing high-precision positioning of the entrance corner point, enhancing the accuracy and reliability of the entrance corner point positioning, and laying a foundation for improving the accuracy of the target parking space identification result; further, based on the second position information corresponding to the entrance corner point, the storage space direction and the storage space frame information, the target parking space can be accurately determined, and at the same time, the position information of the storage space tail corner point corresponding to the target parking space can be accurately calculated, thereby avoiding the problem in the prior art that the storage space is identified by relying on four corner points, and the target parking space cannot be accurately identified when the storage space tail corner point is blocked or blurred, thereby effectively improving the accuracy and reliability of storage space identification. Secondly, by adding a self-calibrated convolutional layer to the feature extraction network, it is possible to adaptively establish long-range and inter-channel dependencies around each spatial position during the feature extraction process, so as to explicitly merge rich information to generate discriminative representations, thereby making the feature extraction network more representative. Thirdly, based on the multi-box classification method, the directional regression problem is transformed into a directional classification problem, avoiding the problem of poor accuracy in location direction recognition due to direct directional regression in the prior art, and improving the accuracy of location direction recognition.

[0159] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0160] Based on the same inventive concept, the embodiment of the present application also provides a parking space identification device for implementing the parking space identification method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more parking space identification device embodiments provided below can refer to the above limitations on the parking space identification method, and will not be repeated here.

[0161] In an exemplary embodiment, Fig.11 As shown, a parking space identification device is provided, comprising: an acquisition module 1101, an identification module 1102, a position information correction module 1103 and a parking space determination module 1104, wherein:

[0162] An acquisition module 1101 is used to acquire a storage location feature map and a key point heat map of the storage location image to be processed;

[0163] The identification module 1102 is used to predict the storage location feature map to obtain an initial storage location identification result; wherein the initial storage location identification result includes the first position information of the entrance corner point corresponding to the storage location, the storage location direction and the storage location frame information;

[0164] The position information correction module 1103 is used to determine the target key point that matches the first position information of the entry corner point from the key point heat map, and determine the position information of the target key point as the second position information of the entry corner point; the accuracy of the second position information is greater than that of the first position information;

[0165] The storage location determination module 1104 is used to determine the target parking space according to the second position information of the entrance corner point, the storage location direction and the storage location frame information.

[0166] The above-mentioned parking space identification device, based on the storage space feature map of the storage space image to be processed, realizes preliminary prediction of the storage space image to be processed from a global perspective, and based on this, it can quickly locate the potential storage space area; further, based on the key point heat map corresponding to the storage space image to be processed and the first position information of the entrance corner point, it can accurately match the second position information corresponding to the entrance corner point, and the accuracy of the second position information is greater than the first position information, thereby realizing high-precision positioning of the entrance corner point, enhancing the accuracy and reliability of the entrance corner point positioning, and laying a foundation for improving the accuracy of the target parking space identification result; further, based on the second position information corresponding to the entrance corner point, the storage space direction and the storage space frame information, the target parking space can be accurately determined, and at the same time, the position information of the storage space tail corner point corresponding to the target parking space can be accurately calculated, avoiding the problem in the prior art that the storage space is identified by relying on four corner points, and the target parking space cannot be accurately identified when the storage space tail corner point is blocked or blurred, thereby effectively improving the accuracy and reliability of storage space identification.

[0167] In one embodiment, the acquisition module 1101 is further used to

[0168] Get the image of the storage location to be processed;

[0169] The image of the storage location to be processed is input into the trained parking space detection model, and feature extraction is performed through the feature extraction network in the trained parking space detection model to generate a storage location feature map and a key point heat map; wherein the feature extraction network includes a self-calibration convolutional layer.

[0170] In one embodiment, the storage location feature map includes storage location feature information; the identification module 1102 is also used to

[0171] The storage location feature map is input into the prediction network in the trained parking space detection model, and the storage location feature information is predicted by the prediction network to obtain the initial storage location recognition result.

[0172] In one embodiment, the parking space recognition device further includes a model training module;

[0173] Model training module for

[0174] Obtain multiple annotated sample parking space bird's-eye views and an initial parking space detection model;

[0175] Construct the target loss function based on the labeled sample storage location bird's-eye view;

[0176] According to the labeled sample parking space bird's-eye view, the initial parking space detection model is trained until the target loss function meets the preset convergence condition, and a trained parking space detection model is obtained.

[0177] In one embodiment, the model training module is also used to

[0178] Input the labeled sample storage location bird's-eye view into the initial parking space detection model to obtain the sample storage location recognition result;

[0179] According to the sample storage location recognition results and the data labels corresponding to the annotated sample storage location bird's-eye view, the storage location entrance corner loss function, storage location direction loss function, storage location frame loss function, storage location depth loss function, storage location category loss function and limiter loss function are constructed respectively;

[0180] The target loss function is obtained by performing weighted summation on the storage location entrance corner point loss function, storage location direction loss function, storage location frame loss function, storage location depth loss function, storage location category loss function and limiter loss function.

[0181] In one embodiment, the data labels include heat map labels; the model training module is also used to

[0182] According to the heat map label and the confidence of the predicted point in the prediction result of the storage location entrance corner point, the storage location entrance corner point loss function is determined.

[0183] In one embodiment, the data label includes a storage location direction label; the model training module is also used to

[0184] Determine a confidence loss function according to the approximate direction angle true value and the angle offset true value in the storage location direction label and the approximate direction angle prediction value and the angle offset prediction value in the storage location direction prediction result;

[0185] Determine the positioning loss function according to the true value of the angle offset in the storage location direction label and the predicted value of the angle offset in the storage location direction prediction result;

[0186] The confidence loss function and the positioning loss function are weightedly summed to obtain the storage location direction loss function.

[0187] In one embodiment, the position information correction module 1103 is also used to

[0188] Based on the preset position mapping relationship and the first position information of the entry corner point, a target key point matching the first position information is determined from the key point heat map.

[0189] Each module in the above parking space identification device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0190] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.12 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store parking space identification related data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a parking space identification method is implemented.

[0191] Those skilled in the art will understand that Fig.12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0192] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0193] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0194] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0195] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0196] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0197] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A parking space identification method, characterized in that: The method comprises: Obtaining a storage location feature map and a key point heat map of the storage location image to be processed; the key point heat map obeys the Lablas distribution; The storage location feature map is predicted to obtain an initial storage location recognition result; wherein the initial storage location recognition result includes the first position information of the entrance corner point corresponding to the storage location, the storage location direction and the storage location frame information; the storage location direction refers to the direction from the entrance corner point to the interior of the parking lot along the entrance corner point; the storage location direction is represented by an approximate direction angle and an angle offset; the approximate direction angle refers to the direction of the storage location based on the multi-box classification method. The division obtained The center angle of the box corresponding to the storage location direction in the boxes; Determine a target key point that matches the first position information of the entry corner point from the key point heat map, and determine the position information of the target key point as the second position information of the entry corner point; the accuracy of the second position information is greater than the first position information; the target key point refers to the key point in the key point heat map with the highest correlation with the entry corner point; Determine a target parking space according to the second position information of the entrance corner point, the storage location direction and the storage location frame information; The determining, from the key point heat map, a target key point that matches the first position information of the entry corner point comprises: Based on a preset position mapping relationship and the first position information of the entry corner point, a target key point matching the first position information is determined from the key point heat map.

2. The method according to claim 1, characterized in that The step of obtaining the storage location feature map and key point heat map of the storage location image to be processed includes: Get the image of the storage location to be processed; The storage location image to be processed is input into a trained parking space detection model, and feature extraction is performed through a feature extraction network in the trained parking space detection model to generate a storage location feature map and a key point heat map; wherein the feature extraction network includes a self-calibration convolution layer.

3. The method according to claim 1, characterized in that The storage location feature map includes storage location feature information; the predicting of the storage location feature map to obtain an initial storage location identification result includes: The storage location feature map is input into the prediction network in the trained parking space detection model, and the storage location feature information is predicted by the prediction network to obtain an initial storage location recognition result.

4. The method according to claim 2, characterized in that: The method further comprises: Obtain multiple annotated sample parking space bird's-eye views and an initial parking space detection model; Constructing a target loss function according to the labeled sample storage location bird's-eye view; According to the labeled sample parking space bird's-eye view, the initial parking space detection model is trained until the target loss function meets a preset convergence condition, thereby obtaining a trained parking space detection model; The constructing a target loss function according to the labeled sample storage location bird's-eye view includes: Inputting the labeled sample storage location bird's-eye view into the initial parking space detection model to obtain a sample storage location recognition result; According to the sample storage location identification results and the data labels corresponding to the labeled sample storage location bird's-eye view, a storage location entry corner point loss function, a storage location direction loss function, a storage location frame loss function, a storage location depth loss function, a storage location category loss function and a limiter loss function are respectively constructed; The target loss function is obtained by performing a weighted sum calculation on the storage location entrance corner point loss function, the storage location direction loss function, the storage location frame loss function, the storage location depth loss function, the storage location category loss function and the limiter loss function.

5. The method according to claim 4, characterized in that The data label includes a heat map label; the sample storage location identification result includes a storage location entrance corner point prediction result; constructing a storage location entrance corner point loss function includes: The storage location entrance corner point loss function is determined according to the heat map label and the prediction point confidence in the storage location entrance corner point prediction result.

6. The method according to claim 4, characterized in that The data label includes a storage location direction label; the sample storage location identification result includes a storage location direction prediction result; and constructing a storage location direction loss function includes: Determine a confidence loss function according to the approximate direction angle true value and the angle offset true value in the storage location direction label, and the approximate direction angle prediction value and the angle offset prediction value in the storage location direction prediction result; Determine a positioning loss function according to the true value of the angle offset in the storage location direction label and the predicted value of the angle offset in the storage location direction prediction result; The confidence loss function and the positioning loss function are weightedly summed to obtain a storage location direction loss function.

7. A parking space identification device, characterized in that: The device comprises: An acquisition module, used to acquire a storage location feature map and a key point heat map of the storage location image to be processed; the key point heat map obeys the Lablas distribution; The identification module is used to predict the storage location feature map to obtain an initial storage location identification result; wherein the initial storage location identification result includes the first position information of the entrance corner point corresponding to the storage location, the storage location direction and the storage location frame information; the storage location direction refers to the direction from the entrance corner point to the interior of the parking garage along the entrance corner point; the storage location direction is represented by an approximate direction angle and an angle offset; the approximate direction angle refers to the first position information of the entrance corner point corresponding to the storage location, the storage location direction and the storage location frame information; the storage location direction refers to the direction of the parking garage from the entrance corner point to the entrance corner point; the storage location direction is represented by an approximate direction angle and an angle offset; the approximate direction angle refers to the direction of the parking garage from the entrance corner point to the entrance corner point based on the multi-box classification method. The division obtained The center angle of the box corresponding to the storage location direction in the boxes; a position information correction module, for determining a target key point that matches the first position information of the entry corner point from the key point heat map, and determining the position information of the target key point as the second position information of the entry corner point; the accuracy of the second position information is greater than that of the first position information; the target key point refers to the key point in the key point heat map with the highest correlation with the entry corner point; A storage location determination module, used for determining a target parking space according to the second position information of the entrance corner point, the storage location direction and the storage location frame information; The position information correction module is also used to determine the target key point matching the first position information from the key point heat map based on a preset position mapping relationship and the first position information of the entry corner point.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images