Parking space automatic distribution method and system based on deep learning

Abstract

The invention relates to the technical field of traffic systems, and discloses a parking space automatic distribution method and system based on deep learning, and the method comprises the steps: obtaining the vehicle data of a to-be-distributed parking space and the parking lot data; the acquired vehicle data are input into a preset parking space automatic distribution model, and the parking space automatic distribution model comprises a door control sub-model and a plurality of data processing sub-models; according to the obtained vehicle data, the vehicle size is analyzed through a door control sub-model, and the vehicle type is determined according to the vehicle size and a preset vehicle type judgment standard; selecting a corresponding data processing sub-model according to the determined vehicle type; and according to the acquired vehicle data and parking lot data, data analysis is performed through the selected data processing sub-model, and the allocated parking space is determined. The collaborative architecture of the door control sub-model and the special data processing sub-model realizes rapid classification of vehicle types and accurate matching of parking spaces.

Description

Technical Field

[0001] This application relates to the technical field of transportation systems, and particularly to an automatic parking space allocation method and system based on deep learning. Background Art

[0002] With the acceleration of the urbanization process and the continuous increase in the number of automobiles, the problem of difficult urban parking has become increasingly prominent. The number and scale of various parking lots are also constantly expanding. The contradiction between the limited space resources of parking lots and the increasing number of vehicles makes the management of parking lots a difficult problem in urban traffic management. Traditional parking management systems usually allocate parking spaces by manual allocation or simple rule matching. This method is not only inefficient, prone to problems such as waste of parking spaces and difficulty in finding vehicles, but also unable to meet the personalized needs of users.

[0003] The existing technology mainly collects parking space images through cameras, and then uses image processing technology to identify vehicles and parking spaces. This technology can improve the management efficiency of parking lots to a certain extent, but there are still some problems.

[0004] Firstly, due to the limitations of image processing technology itself, its accuracy and efficiency are restricted to a certain extent. Secondly, this technology requires a large amount of computing resources and storage space, and for large-scale parking lots, the cost is relatively high. In addition, the existing technology also has problems such as single function and low intelligence in vehicle finding and parking lot lighting control. Therefore, how to achieve the intelligent management of parking lots, realize real-time intelligent management, improve parking efficiency, and reduce carbon emissions has become an important topic in the current traffic management field. Summary of the Invention

[0005] This application provides an automatic parking space allocation method and system based on deep learning, which realizes the automatic allocation of parking spaces through a trained automatic parking space model, and is used to solve problems such as uneven allocation of parking spaces and low parking efficiency in the prior art.

[0006] In a first aspect, this application provides an automatic parking space allocation method based on deep learning. The automatic parking space allocation method includes: obtaining vehicle data and parking lot data of the parking space to be allocated; inputting the obtained vehicle data into a preset automatic parking space allocation model, where the automatic parking space allocation model includes a gating sub-model and multiple data processing sub-models; analyzing the vehicle size according to the obtained vehicle data through the gating sub-model, and determining the vehicle type according to the vehicle size and a preset vehicle type judgment standard; selecting a corresponding data processing sub-model according to the determined vehicle type; and performing data analysis according to the obtained vehicle data and parking lot data through the selected data processing sub-model to determine the allocated parking space.

[0007] Optionally, before obtaining the vehicle data and parking lot data of the parking space to be allocated, the automatic parking space allocation method further includes: dividing the obtained historical vehicle data and parking lot data into a training data set, a validation data set, and a test data set; using the training data set to train the automatic parking space allocation model; during the training process, using the validation data set to perform performance verification on the automatic parking space allocation model; and inputting the test data set into the trained automatic parking space allocation model for testing.

[0008] Optionally, the determining the allocated parking space by performing data analysis through the selected data processing sub-model according to the obtained vehicle data and parking lot data includes: estimating the vehicle parking duration according to the obtained vehicle data; determining the size information of the vacant parking spaces and the parking lot charging standard according to the obtained parking lot data; calculating the parking fees for different parking spaces according to the estimated parking duration and the determined parking lot charging standard; selecting a suitable parking space according to the determined size information of the vacant parking spaces and the vehicle type; screening out the parking spaces with lower fees among the selected parking spaces according to the calculated fees; calculating the handling path of the AGV handling vehicle to each of the screened parking spaces; and selecting the parking space corresponding to the shortest path in the handling path as the optimal parking space.

[0009] Optionally, the calculating the time cost of the AGV handling vehicle to each of the screened parking spaces according to the preset path judgment algorithm includes: obtaining all the real-time positions of the AGVs and the parking lot map data; generating a handling path according to the parking lot map data and the positions of each of the screened parking spaces; predicting the future driving paths of other AGVs according to the obtained real-time positions of the AGVs by using the preset path prediction algorithm; adjusting the generated handling path according to the predicted driving paths; and selecting the parking space corresponding to the shortest path in the adjusted handling path as the optimal parking space.

[0010] Optionally, the adjusting the generated handling path according to the predicted driving paths includes: generating a corresponding spatio-temporal trajectory sequence according to the predicted driving paths and the generated handling path; determining spatio-temporal intersection points according to the generated spatio-temporal trajectory sequence; and adjusting the handling path corresponding to the determined spatio-temporal intersection points.

[0011] Optionally, training the parking space automatic allocation model using the training data set includes: using the gating sub-model to extract features from the sample data in the training data set to obtain a feature extraction result; according to the feature extraction result, using the gating sub-model to distribute the sample data in the training data set to different data processing sub-models; using the data processing sub-models to respectively perform different types of data processing training according to the received sample data.

[0012] Optionally, during the training process, validating the performance of the parking space automatic allocation model using the validation data set includes the following validation metrics: determining the average accuracy of the gating sub-model and each of the data processing sub-models; calculating the harmonic mean according to the proportion of samples correctly classified as positive among all samples predicted as positive and the proportion of samples correctly classified as positive among all samples actually being positive, so as to evaluate the model performance of the gating sub-model and each of the data processing sub-models according to the harmonic mean; using a loss function to evaluate the training effect of the gating sub-model and each of the data processing sub-models.

[0013] Optionally, inputting the test data set into the trained parking space automatic allocation model for testing includes the following test metrics: testing the average precision of the gating sub-model and each of the data processing sub-models under different intersection over union (IoU) thresholds; testing the harmonic mean of the precision and recall of the gating sub-model and each of the data processing sub-models for processing different types of data; testing the detection performance of the gating sub-model and each of the data processing sub-models for processing different types of data.

[0014] Optionally, after determining the optimal parking space from the selected parking spaces according to the determined size information of the vacant parking space and the vehicle type, the parking space automatic allocation method further includes: collecting feedback information from users to optimize the model parameters of the parking space automatic allocation model according to the feedback information.

[0015] In a second aspect, the present application provides a parking system, the parking system includes a parking lot, an AGV, and a parking space automatic allocation system configured for the parking lot. The parking space automatic allocation system includes a control device, and the control device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the above-mentioned parking space automatic allocation method.

[0016] The above-mentioned at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: In the technical solution provided by this application, by preprocessing historical parking data and parking lot data, a training set, a validation data set, and a test data set are obtained. The training model is trained using the training data set. During the training process, the validation data set is used to verify the model performance of the training model, and a trained parking space automatic allocation model is obtained. Then, the test data set is input into the parking space automatic allocation model for model testing. If the test result meets the preset standard, the parking space automatic allocation model can be used to automatically allocate parking spaces.

[0017] In this way, the trained parking space automatic allocation system can, according to different tasks and data processing requirements through the gating sub-model, select different data processing sub-models to perform different data processing, dynamically allocate tasks to the most suitable sub-models. In this way, each data processing does not require mobilizing the entire model's calculation, but only needs to call several required data processing sub-models, which can improve the processing efficiency of the model, and also improve the overall performance and generalization ability of the model. During the process of automatic parking space allocation, the optimal parking space for the user can be quickly allocated, improving the efficiency of parking space management, making the parking lot management more intelligent, and effectively improving the user experience. Brief Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a flowchart of a method for automatically allocating parking spaces based on deep learning provided in an embodiment of this application; Figure 2 It is a schematic diagram of the overall process of a method for automatically allocating parking spaces based on deep learning provided in an embodiment of this application; Figure 3 It is a schematic diagram of the model training process provided in an embodiment of this application; Figure 4 It is a logic flowchart of parking space allocation provided in an embodiment of this application; Figure 5 It is a flowchart of AGV path optimization provided in an embodiment of this application; Figure 6 It is a schematic diagram of the specific application process of a method for automatically allocating parking spaces based on deep learning provided in an embodiment of this application. Detailed Embodiments

[0020] The embodiments of the present application provide a method and system for automatic parking space allocation based on deep learning. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims, and the above-mentioned drawings of the present application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances, so that the embodiments described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0021] To make the purpose, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present application.

[0022] Please refer to Figures 1-6 , for ease of understanding, the specific process of the embodiments of the present application will be described below.

[0023] As Figure 1 and Figure 2 shown, the embodiments of the present invention provide a method for automatic parking space allocation based on deep learning. The method for automatic parking space allocation includes the following steps: Step 101: Obtain vehicle data and parking lot data of the parking space to be allocated; Step 102: Input the obtained vehicle data into a preset automatic parking space allocation model, where the automatic parking space allocation model includes a gating sub-model and multiple data processing sub-models; Step 103: Analyze the vehicle size through the gating sub-model according to the obtained vehicle data, and determine the vehicle type according to the vehicle size and a preset vehicle type judgment criterion; Step 104: Select the corresponding data processing sub-model according to the determined vehicle type; Step 105: Perform data analysis according to the obtained vehicle data and parking lot data through the selected data processing sub-model to determine the allocated parking space.

[0024] Among them, the gating sub-model is responsible for dynamically selecting which data processing sub-models to call according to the input data. The module sub-model can specifically be a small neural network, which indicates whether the data processing sub-model is activated by outputting the identifier of each data processing sub-model and the corresponding weight vector. If it is activated, it receives relevant data and performs data processing.

[0025] The gating sub-model extracts features from the input data, extracts features related to the data processing task. In this way, it can be determined which relevant data processing sub-models need to be called according to the extracted data features. The gating sub-model can learn the feature distributions of different data processing tasks through training, so as to realize the dynamic allocation of various different data processing tasks.

[0026] Each data processing sub-model is an independent neural network model. In the actual application process, only some of the required data processing sub-models will be called during each data processing process, rather than the entire large model, which can ensure the efficiency of calculation.

[0027] For example, in a specific application scenario, the data processing sub-module can include a license plate recognition sub-model, a vehicle feature recognition sub-model, a parking lot environment data processing sub-model, a parking space management sub-model, a parking space allocation sub-model, a user feedback sub-model, etc., and can be divided differently according to actual needs, which will not be specifically limited here.

[0028] Such as Figure 3 As shown, preferably, before obtaining the vehicle data and parking lot data of the to-be-allocated parking space, the parking space automatic allocation method further includes: dividing the obtained historical vehicle data and parking lot data into a training data set, a validation data set, and a test data set; using the training data set to train the parking space automatic allocation model; during the training process, using the validation data set to verify the performance of the parking space automatic allocation model; and inputting the test data set into the trained parking space automatic allocation model for testing.

[0029] In a preferred embodiment of the present invention, after obtaining the historical vehicle data and parking lot data, cleaning and normalization operations can also be performed on the historical parking data and the parking lot data.

[0030] In a preferred embodiment of the present invention, the historical data may specifically refer to the historical data related to user parking in a parking lot, such as parking location, parking time, license plate information, vehicle feature information, user type, etc., which are not specifically limited herein. The parking lot data may specifically be historical information such as parking space information and environmental information of the parking lot, which are not specifically limited herein. By collecting historical parking data and parking lot data, actual data can be used to train the training model, so as to facilitate the training model to learn the specific implementation manner of parking space allocation more quickly and accurately, and improve the accuracy of model implementation.

[0031] In a specific application scenario, data collection can be carried out through devices such as geomagnetic sensors, ultrasonic sensors, cameras, etc. to collect parking space information, vehicle information, environmental information, etc. of the parking lot, or historical data stored in the database of the parking lot can be obtained, which is not specifically limited herein.

[0032] In a preferred embodiment of the present invention, the training data set is used for training the parking space automatic allocation model, and the training data set accounts for 70%-80% of the data set. The validation data set is used to regularly evaluate and verify the performance of the model during the training process, and the validation data set accounts for 10%-15% of the data set. The test data set is used to evaluate and judge the final performance of the model after the model training is completed, so as to determine whether the trained parking space automatic allocation model can be put into practical application, and the test data set accounts for 10%-15% of the data set.

[0033] As Figure 4 shown, preferably, the determination of the allocated parking space by performing data analysis through the selected data processing sub-model according to the obtained vehicle data and parking lot data includes: estimating the vehicle parking duration according to the obtained vehicle data; determining the size information of the available parking spaces and the parking lot charging standard according to the obtained parking lot data; calculating the parking fees for different parking spaces according to the estimated parking duration and the determined parking lot charging standard; selecting a suitable parking space according to the determined size information of the available parking spaces and the vehicle type; screening out the parking spaces with lower fees among the selected parking spaces according to the calculated fees; calculating the handling paths of the AGV handling vehicle to each of the screened parking spaces; and selecting the parking space corresponding to the shortest path among the handling paths as the optimal parking space.

[0034] In a specific application scenario, the data processing sub-model uses vehicle data such as the vehicle's usage history and user input information, and applies algorithms such as time series analysis to estimate the vehicle's parking duration. At the same time, based on the real-time monitoring of the parking lot sensors and the pre-entered layout information and other parking lot data, the size information of the vacant parking spaces is determined, including length, width, etc., and the charging standards for different areas and different types of parking spaces in the parking lot are mastered. Then, the size information of the vacant parking spaces is compared with the vehicle type determined by the gating sub-model. If the vehicle is a large truck, select a parking space with a large enough size and an open surrounding space for the truck to enter and exit as the optimal parking space from the vacant parking spaces; if it is an ordinary car, select a parking space with a suitable size and a convenient location. Subsequently, the estimated parking duration is combined with the determined charging standard, and the cost calculation model is used to calculate the parking fees required for different parking spaces. Based on the cost calculation results, preferentially screen out the parking spaces whose fees are within the acceptable range of the user or the most economical ones. Finally, calculate the handling path of the AGV handling vehicle to each screened parking space, and select the parking space corresponding to the shortest path as the optimal parking space, thus completing the process of determining the parking space and realizing an efficient, accurate and user-demand-fitting parking space allocation.

[0035] As Figure 5 shown, preferably, calculating the time cost of the AGV handling vehicle to each screened parking space according to the preset path judgment algorithm includes: obtaining all the real-time positions of the AGVs and the parking lot map data; generating a handling path according to the parking lot map data and the positions of each screened parking space; predicting the future driving paths of other AGVs using the preset path prediction algorithm according to the obtained real-time positions of the AGVs; adjusting the generated handling path according to the predicted driving paths; and selecting the parking space corresponding to the shortest path among the adjusted handling paths as the optimal parking space.

[0036] In a specific application scenario, when calculating the time cost of the AGV handling vehicle to each screened parking space, first, obtain all the real-time positions of the AGVs and the parking lot map data to provide basic information for subsequent path planning. Generate a preliminary handling path based on the parking lot map and the positions of the screened parking spaces. At the same time, use a preset algorithm (such as the Dijkstra algorithm) to predict the future driving paths of other AGVs according to the real-time positions of the AGVs, which takes into account the dynamics of the AGV operation in the parking lot. Then, adjust the preliminarily generated handling path according to the prediction results to avoid path conflicts and improve the overall operation efficiency. Finally, select the parking space corresponding to the shortest path from the adjusted paths as the optimal parking space, which not only ensures the AGV handling efficiency, reduces the operation time and energy consumption, but also improves the overall operation efficiency of the parking lot and the user's parking experience, and improves the parking space allocation strategy based on AGV handling.

[0037] Preferably, adjusting the generated handling path according to the predicted driving path includes: generating a corresponding spatio-temporal trajectory sequence according to the predicted driving path and the generated handling path; determining spatio-temporal intersection points according to the generated spatio-temporal trajectory sequence; and adjusting the handling path corresponding to the determined spatio-temporal intersection points.

[0038] In a specific application scenario, when an initial handling path of a certain vehicle is generated based on the real-time position of the AGV and the parking lot map data, and the future driving paths of other AGVs are obtained by using a preset path prediction algorithm, the path adjustment process will be started. The system first converts the predicted path and the already generated handling path into spatio-temporal trajectory sequences, which detail the spatial positions of the AGV at different times. Then, the system compares the spatio-temporal trajectory sequences of each AGV to find the spatio-temporal intersection points, which represent that different AGVs will be in the same spatial position at the same time, possibly causing collisions or mutual hindrances. Once the spatio-temporal intersection points are determined, the system re-plans the handling paths of the AGVs at risk of conflict, calculates new paths that can avoid the intersection points, and finally enables the AGV to efficiently and safely complete the vehicle handling task along the adjusted path, ensuring the stable operation of the parking lot handling system.

[0039] For example, a private car drives towards the entrance of a parking lot. Cameras and sensors quickly capture vehicle information, including license plate numbers, vehicle type outlines, etc. By comparing with the vehicle management database, the system finds that when this car parked here on weekdays in the past, the average parking duration was about 7 - 8 hours, and thus estimates that the parking duration this time is also within this range. Meanwhile, the parking lot management system updates the data in real time, showing that there are multiple vacant standard small parking spaces near the main entrance of the shopping mall. These parking spaces are 5.2 meters long, 2.3 meters wide, and charge 8 yuan per hour, suitable for small cars to park. In the relatively remote ordinary parking space area of the parking lot, the parking space size is similar, but the charge is 6 yuan per hour. Considering the estimated parking duration of this car and the charging standards in different areas, the system calculates that the parking fee for 8 hours in the standard parking space area near the main entrance of the shopping mall is 64 yuan, and the parking fee for 8 hours in the ordinary parking space area is 48 yuan. Since the owner's past records show that he is sensitive to prices, the system preferentially allocates the parking space in the ordinary parking space area to this private car. After the vehicle enters the parking lot, since AGV is used for vehicle handling in some areas of this parking lot, the system obtains the real-time positions of all AGVs and the parking lot map data, and generates an initial handling path for this private car according to the position of the allocated parking space. However, using the preset path prediction algorithm, the system finds that the future driving path of another running AGV may conflict with the initial handling path at a certain space-time point. So, the system adjusts the initial handling path according to the prediction result and re-plans a new path to avoid the conflict. Finally, the AGV transports the private car to the designated ordinary parking space quickly along the adjusted shortest path. The owner completes parking in a short time and goes to the shopping mall to work smoothly. The whole process not only realizes the collaborative utilization of vehicle and parking lot data, but also determines a suitable parking space through multi-dimensional screening. At the same time, with the help of the optimization decision of the AGV handling path, the parking operation is completed efficiently and safely, improving the overall operation efficiency of the parking lot and meeting the parking needs of the owner.

[0040] Preferably, using the training data set to train the parking space automatic allocation model may specifically include: using the gating sub-model to extract features from the sample data in the training data set to obtain a feature extraction result; according to the feature extraction result, using the gating sub-model to distribute the sample data in the training data set to different data processing sub-models; using the data processing sub-models to perform different types of data processing training according to the received sample data.

[0041] Specifically, the gating sub-model can extract feature data from the sample data in the training dataset to obtain task features related to data processing. For example, for vehicle type recognition, data features related to the vehicle type, such as length, width, and height, can be extracted. For environmental information processing, features related to the parking lot environment, such as temperature, humidity, and brightness, can be extracted. Specific limitations are not provided here. After the gating sub-model extracts the data features, different sample data can be assigned to different data processing sub-models for processing according to the data features. For example, sample data related to sedans can be assigned to the sedan data processing sub-model.

[0042] Further preferably, the method may further include: during the process of training the training model using the training dataset, optimizing the model parameters through the backpropagation algorithm. Among them, the backpropagation algorithm starts from the output layer and calculates the gradients of each hierarchical architecture in the model layer by layer in reverse.

[0043] Preferably, using the validation dataset to verify the performance of the parking space automatic allocation model may include the following verification metrics: determining the average accuracy of the gating sub-model and each of the data processing sub-models; calculating the harmonic mean based on the proportion of samples correctly classified as positive among all samples predicted as positive and the proportion of samples correctly classified as positive among all samples actually being positive, so as to evaluate the model performance of the gating sub-model and each of the data processing sub-models according to the harmonic mean; using the loss function to evaluate the training effect of the gating sub-model and each of the data processing sub-models.

[0044] Specifically, by regularly using the validation dataset to verify the model performance of the training model during the model training process, the model parameters can be adjusted to prevent overfitting.

[0045] Among them, the loss function may include a classification loss function, a regression loss function, and a regularization loss function, etc. The classification loss function is used to evaluate the performance of the model in classification tasks, such as the parking space allocation task and the vehicle recognition task, etc.; the regression loss function is used to evaluate the performance of the model in regression tasks, such as the distance and angle of parking space allocation, etc.; the regularization loss function is used to prevent the model from overfitting and improve the generalization ability of the model.

[0046] Preferably, inputting the test dataset into the trained parking space automatic allocation model for testing may include the following test metrics: testing the average precision of the gating sub-model and each of the data processing sub-models under different intersection over union (IoU) thresholds; testing the harmonic mean of the precision and recall of the gating sub-model and each of the data processing sub-models for processing different types of data; testing the detection performance of the gating sub-model and each of the data processing sub-models for processing different types of data.

[0047] Further preferably, after determining the optimal parking space from the selected parking spaces according to the determined size information of the vacant parking spaces and the vehicle type, the parking space automatic allocation method further includes: collecting feedback information of the user to optimize the model parameters of the parking space automatic allocation model according to the feedback information.

[0048] In the embodiments of the present specification, after the parking space automatic allocation model is put into use, feedback information of the user on the parking space allocation result can be collected to view the satisfaction degree of the user on the allocation result, and the user feedback data can be further analyzed in depth to mine the needs and preferences of the user, predict the behavior pattern of the user, and further adjust the parameters of the parking space automatic allocation model, such as the learning rate, regularization coefficient, etc., to improve the accuracy and generalization ability of the model.

[0049] In addition, the user feedback information data can also be used as new training samples to update the parking space automatic allocation model regularly, so that the model can adapt to the changes of user needs and the dynamic changes of the parking lot environment, further optimize the model algorithm, and improve the adaptability, intelligence level, accuracy and efficiency of parking space allocation, and improve the user experience.

[0050] Based on the same inventive concept, as Figure 6 shown, it is a schematic flow chart of the actual application of a parking space automatic allocation method based on deep learning provided in the embodiments of the present application.

[0051] Among them, the actual application steps of the parking space automatic allocation method based on deep learning are as follows: Step S302: Collect the parking space information, environmental information and vehicle to be parked information of the parking lot; Step S304: Perform preprocessing operations such as cleaning and normalization on the collected data to obtain the processed target data; Step S306: Input the target data into the parking space automatic allocation model; Step S308: The parking space automatic allocation model outputs a parking space allocation plan; Step S310: The parking lot handling robot transports the target vehicle to the designated parking space according to the parking space allocation plan; Step S312: Receive user feedback information and update the parking space automatic allocation model according to the feedback information.

[0052] The parking space automatic allocation method based on deep learning in the embodiments of the present application has been described above. Next, the parking space automatic allocation system based on deep learning in the embodiments of the present application will be described. Please refer to Figure 5 which is a schematic structural diagram of a parking space automatic allocation system based on deep learning provided by the present application.

[0053] The present invention also provides a parking system, which includes a parking lot, an AGV, and a parking space automatic allocation system configured for the parking lot. The parking space automatic allocation system includes a control device, and the control device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the above-mentioned parking space automatic allocation method.

[0054] A parking space automatic allocation method and system based on deep learning provided by an embodiment of this specification preprocess historical parking data and parking lot data to obtain a training set, a validation data set, and a test data set. The training data set is used to train a training model. During the training process, the validation data set is used to verify the model performance of the training model to obtain a trained parking space automatic allocation model. The automatic allocation model realizes intelligent decision-making for parking space allocation through the collaborative architecture of a gating sub-model and a multi-data processing sub-model, combined with a dynamic cost calculation, vehicle type adaptive matching, and user feedback closed-loop optimization mechanism.

[0055] Meanwhile, the gating sub-model can select different data processing sub-models according to different tasks and data processing requirements to perform different data processing, and dynamically allocate tasks to the most suitable sub-model. In this way, each data processing does not require mobilizing the calculation of the entire model, but only needs to call several required data processing sub-models, which can improve the processing efficiency of the model, and also improve the overall performance and generalization ability of the model. During the process of automatic parking space allocation, the optimal parking space for the user can be quickly allocated, improving the efficiency of parking space management, making the parking lot management more intelligent, and effectively improving the user experience.

[0056] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0057] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment.

[0058] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for automatic parking space allocation based on deep learning, characterized in that: The automatic parking space allocation method comprises: Obtain vehicle data and parking lot data for parking spaces to be allocated; Inputting the acquired vehicle data into a preset automatic parking space allocation model, wherein the automatic parking space allocation model includes a gate control sub-model and a plurality of data processing sub-models; Analyzing the vehicle size through the gated sub-model according to the acquired vehicle data, and determining the vehicle type according to the vehicle size and a preset vehicle type judgment standard; According to the determined vehicle type, a corresponding data processing sub-model is selected; According to the acquired vehicle data and parking lot data, data analysis is performed through the selected data processing sub-model to determine the allocated parking space.

2. The automatic parking space allocation method according to claim 1, characterized in that: Before acquiring the vehicle data and parking lot data of the parking space to be allocated, the automatic parking space allocation method further includes: Divide the acquired historical vehicle data and parking lot data into a training data set, a validation data set, and a test data set; Using the training data set, training the automatic parking space allocation model; During the training process, the performance of the automatic parking space allocation model is verified using the verification data set; The test data set is input into the trained automatic parking space allocation model for testing.

3. The automatic parking space allocation method according to claim 1, characterized in that: The method of performing data analysis based on the acquired vehicle data and parking lot data and determining the allocated parking space through the selected data processing sub-model includes: According to the acquired vehicle data, estimate the parking time of the vehicle; Determine the size information of the vacant parking spaces and parking fee standards based on the acquired parking lot data; Calculate the parking fees at different parking spaces based on the estimated parking duration and the determined parking fee standards; Selecting a suitable parking space based on the determined size information of the vacant parking space and the type of vehicle; According to the calculated fee, the selected parking spaces are screened for parking spaces with lower fees; Calculate the transport path of the AGV transport vehicle to each selected parking space; The parking space corresponding to the shortest path in the transport path is selected as the optimal parking space.

4. The automatic parking space allocation method according to claim 3, characterized in that: The time cost of transporting the vehicle to each selected parking space by the AGV is calculated according to the preset path judgment algorithm, including: Get all AGV real-time location and parking lot map data; Generate a transport path based on the parking lot map data and the location of each selected parking space; Based on the real-time position of the AGV obtained, the preset path prediction algorithm is used to predict the future driving paths of other AGVs; Adjusting the generated transport path according to the predicted travel path; The parking space corresponding to the shortest path in the adjusted transport path is selected as the optimal parking space.

5. The automatic parking space allocation method according to claim 4, characterized in that: The step of adjusting the generated transport path according to the predicted travel path includes: Generate a corresponding spatiotemporal trajectory sequence according to the predicted driving path and the generated transport path; Determine the space-time intersection point according to the generated space-time trajectory sequence; The transport path corresponding to the determined time-space intersection point is adjusted.

6. The automatic parking space allocation method according to claim 2, characterized in that: Using the training data set, training the automatic parking space allocation model includes: Using the gated sub-model, extracting features from the sample data in the training data set to obtain feature extraction results; According to the feature extraction result, using the gating sub-model, the sample data in the training data set is distributed to different data processing sub-models; The data processing sub-model is used to perform different types of data processing training according to the received sample data.

7. The automatic parking space allocation method according to claim 2, characterized in that: During the training process, the performance of the automatic parking space allocation model is verified using the verification data set, including the following verification indicators: Determining the average accuracy of the gating sub-model and each of the data processing sub-models; Calculating a harmonic mean according to the proportion of samples correctly classified as positive to all samples predicted to be positive and the proportion of samples correctly classified as positive to all samples actually being positive, so as to evaluate the model performance of the gating sub-model and each of the data processing sub-models according to the harmonic mean; The loss function is used to evaluate the training effects of the gating sub-model and each of the data processing sub-models.

8. The automatic parking space allocation method according to claim 2, characterized in that: The test data set is input into the trained automatic parking space allocation model for testing, including the following test indicators: Testing the average accuracy of the gating sub-model and each of the data processing sub-models at different intersection-over-union ratio thresholds; Testing the harmonic mean of the precision and recall of the gating sub-model and each of the data processing sub-models in processing different types of data; The detection performance of the gating sub-model and each of the data processing sub-models in processing different types of data is tested.

9. The automatic parking space allocation method according to claim 1, characterized in that: After determining the optimal parking space from the selected parking spaces according to the determined size information of the vacant parking spaces and the vehicle type, the automatic parking space allocation method further includes: Collect user feedback information to optimize model parameters of the automatic parking space allocation model according to the feedback information.

10. A parking system, characterized in that: The parking system includes a parking lot, an AGV, and an automatic parking space allocation system configured in the parking lot. The automatic parking space allocation system includes a control device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the automatic parking space allocation method according to any one of claims 1-9.

Images