A high-position video monitoring method and system for a parking lot based on deep recognition

Abstract

The application discloses a kind of high-position video monitoring method and system of parking lot based on depth recognition, belong to intelligent transportation field, wherein method includes: constructing high-position camera group, vehicle is identified and tracked, obtains vehicle information;Real-time parking space information is obtained in parking lot;Vehicle database stores the historical parking information of each vehicle.When target vehicle enters, find the vehicle information in vehicle database, if exists, according to its historical parking information and real-time parking space information, it is intelligently allocated parking space;If not, obtain its destination information by interacting with the car owner, according to destination information and real-time parking space information, it is allocated parking space.Vehicle leaves, and its leaving information is confirmed by exit camera, and vehicle database is updated.The present application solves the technical problem that vehicle cannot be accurately identified and intelligently managed in the prior art, achieves the technical effect of parking lot video full monitoring and vehicle intelligent parking space allocation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent transportation, and particularly relates to a high-position video monitoring method and system for parking lots based on deep recognition. BACKGROUND

[0002] With the acceleration of urbanization and the rapid increase in the number of vehicles, the shortage of parking spaces has become a major problem in urban transportation. Most of the existing parking lot management systems use manual management methods, which have low vehicle management efficiency and unreasonable vehicle parking, and thus cannot efficiently monitor and manage vehicles. SUMMARY

[0003] The present application provides a high-position video monitoring method and system for parking lots based on deep recognition, aiming to solve the technical problem that the prior art cannot accurately identify and intelligently manage vehicles.

[0004] In view of the above problems, the present application provides a high-position video monitoring method and system for parking lots based on deep recognition.

[0005] The first aspect of the present application provides a high-position video monitoring method for parking lots based on deep recognition, which comprises: constructing a high-position camera group, the high-position camera group comprising an entrance camera, a panoramic camera and an exit camera; photographing a target vehicle through the entrance camera to obtain a target vehicle image; photographing the parking lot through the panoramic camera to obtain a panoramic image of the parking lot, and obtaining parking distribution information according to the panoramic image of the parking lot; inputting the target vehicle image into a target detection model to obtain target vehicle information; traversing a vehicle database to obtain first vehicle information and matching the target vehicle information; when the target vehicle information matches the first vehicle information, obtaining target vehicle historical parking information according to the first vehicle information, and assigning a parking space to the target vehicle according to the target vehicle historical parking information and the parking distribution information; when the target vehicle information does not match the first vehicle information, adding the target vehicle to the vehicle database, interacting with the target vehicle, obtaining the destination of the target vehicle, and assigning a parking space to the target vehicle according to the destination and the parking distribution information; obtaining exit information of the target vehicle through the exit camera, and adding the exit to the vehicle database.

[0006] Another aspect of the present application provides a high-position video monitoring system for parking lots based on deep recognition, comprising: a camera group construction module for constructing a high-position camera group, the high-position camera group comprising an entrance camera, a panoramic camera and an exit camera; a target vehicle image module for capturing a target vehicle through the entrance camera to obtain a target vehicle image; a parking distribution information module for capturing a panoramic image of the parking lot through the panoramic camera to obtain parking distribution information according to the panoramic image of the parking lot; a target vehicle information module for inputting the target vehicle image into a target detection model to obtain target vehicle information; a vehicle information matching module for traversing a vehicle database to obtain first vehicle information and matching the target vehicle information; a vehicle information existence module for, when the target vehicle information matches the first vehicle information, obtaining target vehicle historical parking information according to the first vehicle information and assigning a parking space for the target vehicle according to the target vehicle historical parking information and the parking distribution information; a vehicle information absence module for, when the target vehicle information does not match the first vehicle information, adding the target vehicle to the vehicle database and interacting with the target vehicle to obtain a destination of the target vehicle and assigning a parking space for the target vehicle according to the destination and the parking distribution information; and a vehicle exit information module for obtaining exit information of the target vehicle through the exit camera and adding the exit to the vehicle database.

[0007] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] Since the high-position camera group comprising the entrance camera, the panoramic camera and the exit camera is constructed, the vehicles entering and leaving the parking lot are recognized and tracked by using the deep learning target detection model to obtain vehicle information; at the same time, the real-time parking space information of the parking lot is obtained by the panoramic camera; the historical parking information of each vehicle is stored in the vehicle database; when the target vehicle enters, the vehicle information is first searched in the vehicle database, if the vehicle information exists, a parking space is intelligently assigned for the vehicle according to the historical parking information and the real-time parking space information; if the vehicle information does not exist, the destination information is obtained by interacting with the vehicle owner, and a parking space is also assigned for the vehicle according to the destination information and the real-time parking space information; when the vehicle exits, the exit information is confirmed by the exit camera, and the vehicle database is updated in time, which solves the technical problem that the vehicle cannot be accurately recognized and intelligently managed in the prior art, and achieves the technical effects of full monitoring of the parking lot video and intelligent parking space assignment for the vehicle.

[0009] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A possible flowchart of a high-position video monitoring method for a parking lot based on depth recognition is provided for the embodiments of the present application.

[0011] Figure 2 A possible flowchart of acquiring a target vehicle image in a high-position video monitoring method for a parking lot based on depth recognition is provided for the embodiments of the present application.

[0012] Figure 3 A possible flowchart of acquiring parking distribution information in a high-position video monitoring method for a parking lot based on depth recognition is provided for the embodiments of the present application.

[0013] Figure 4 A possible structure diagram of a high-position video monitoring system for a parking lot based on depth recognition is provided for the embodiments of the present application.

[0014] Legend of reference signs: camera group construction module 11, target vehicle image module 12, parking distribution information module 13, target vehicle information module 14, vehicle information matching module 15, vehicle information existing module 16, vehicle information missing module 17, vehicle off-site information module 18. DETAILED DESCRIPTION

[0015] The general idea of the technical solutions provided by the present application is as follows:

[0016] The embodiments of the present application provide a high-position video monitoring method and system for a parking lot based on depth recognition. High-position camera depth target detection and vehicle database technology are adopted to realize automatic recognition, intelligent management and dynamic allocation of parking spaces for vehicles in the whole process in the parking lot.

[0017] Video images of each vehicle entering and leaving the parking lot are acquired by constructing a high-position camera group, and a depth learning target detection algorithm is used to analyze the video images to identify each vehicle in real time and acquire vehicle information of the vehicle. At the same time, real-time parking space information of the parking lot is monitored by a panoramic camera, and historical parking information of each vehicle is stored in a vehicle database. When a vehicle enters the parking lot, the historical information of the vehicle is first searched in the vehicle database. If the historical information exists, a parking space is intelligently selected for the vehicle according to its historical parking habit and real-time parking space information. If the historical information does not exist, destination information of the vehicle is acquired by interacting with the vehicle owner, and a parking space is allocated for the vehicle according to the destination information and real-time parking space information. When the vehicle leaves, the vehicle is captured again by an off-site camera, it is determined that the vehicle has left, and the vehicle database is updated in time.

[0018] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be specifically introduced in combination with the drawings of the specification.

[0019] Embodiment one

[0020] As Figure 1 shown, the embodiment of the present application provides a high-position video monitoring method for parking lot based on deep recognition, which is applied to a high-position video monitoring system for parking lot, and the method comprises the following steps:

[0021] Step S100: constructing a high-position camera group, wherein the high-position camera group comprises an entrance camera, a panoramic camera and an exit camera.

[0022] Specifically, in order to realize efficient monitoring of the parking lot, a high-position camera group is constructed. First, according to the functional requirements of the entrance camera, the panoramic camera and the exit camera, corresponding camera models are selected, and indicators such as image quality, resolution, viewing angle and protection are considered during selection to ensure that the cameras can stably operate in the outdoor parking lot environment. Second, according to the specific structure and size of the parking lot, the layout of the cameras is designed to ensure that key areas such as the entrance, exit and panoramic view are within the effective monitoring range, avoiding dead angles. Then, according to the layout design, the cameras are installed at appropriate high positions in the parking lot, such as the outer wall or roof of the parking lot, to ensure that the camera's viewing direction and angle meet the design requirements and are firmly installed to prevent movement or tilting. After installation, all cameras are debugged to adjust the focal length and aperture to ensure high image clarity and moderate brightness. The viewing range of each camera is checked to see if there are any obstructions or dead angles. The linkage control of the cameras is tested to ensure synchronous monitoring of the target under different camera angles. Finally, each camera in the high-position camera group is connected to the server of the monitoring center through the network to realize real-time transmission and storage of image information.

[0023] The high-position camera group comprises an entrance camera, a panoramic camera and an exit camera. The entrance camera uses a conventional monitoring camera to obtain image information of a target vehicle entering the parking lot, thereby identifying the identity of the target vehicle and obtaining its historical parking information. The panoramic camera selects a super-wide-angle lens and panoramic stitching technology to take panoramic images of the entire parking lot, obtaining distribution information of vehicles in the parking lot to provide a reference basis for allocating parking spaces for the target vehicle. The selection of the exit camera is similar to that of the entrance camera, which is used to detect and record the off-site information of the target vehicle and add the information to the vehicle database to provide services for the next time the target vehicle enters the parking lot.

[0024] By constructing a camera group comprising an entrance camera, a panoramic camera and an exit camera, the distribution changes of vehicles in the parking lot and the entry and exit states of vehicles are monitored in real time, providing a basis for the automatic monitoring and management of the parking lot.

[0025] Step S200: capturing a target vehicle through the entrance camera to obtain a target vehicle image.

[0026] Specifically, when the target vehicle enters the parking lot, the entrance camera automatically tracks and photographs it to obtain an entrance video stream containing the target vehicle. The entrance camera selects a surveillance camera with license plate recognition function, obtains the license plate information of the target vehicle through license plate recognition technology, and identifies the identity of the target vehicle accordingly.

[0027] After obtaining the entrance video stream, the system processes the video stream to extract the target vehicle image. First, the moving objects in the video stream are detected through frame difference technology, and the vehicle image frames containing the target vehicle are extracted to obtain a vehicle image set. Then, the vehicle image set is screened, and the images with high image quality and containing the target vehicle are selected as the target vehicle images.

[0028] By photographing the target vehicle, the target vehicle image is obtained, effectively capturing the target vehicle image entering the parking lot, providing a basis for identifying the identity of the target vehicle and obtaining its historical parking information, and realizing the automation of parking lot monitoring and management.

[0029] Step S300: capturing the parking lot by the panoramic camera to obtain a panoramic image of the parking lot, and obtaining parking distribution information according to the panoramic image of the parking lot;

[0030] Specifically, the panoramic camera uses a super wide-angle lens and panoramic stitching technology for panoramic shooting to obtain an original panoramic video stream containing the entire field of view of the parking lot. Then, the original video stream is preprocessed, including denoising, brightness adjustment, contrast enhancement, etc., to obtain a set of panoramic images of the parking lot. Then, effective information is extracted from the panoramic images, and all panoramic images are stitched to obtain a composite panoramic image containing the overall view of the parking lot, i.e. the panoramic image of the parking lot.

[0031] A deep learning model such as a convolutional neural network is used to train a large amount of sample data to build a parking distribution recognition model and acquire the ability to recognize the state of each parking space. Based on the panoramic image of the parking lot, a parking distribution recognition model is constructed to analyze the image and obtain information on whether each parking space is occupied.

[0032] By capturing the parking lot by the panoramic camera, the panoramic image of the parking lot is obtained, and the detailed vehicle distribution information in the parking lot is obtained according to the panoramic image of the parking lot, providing an important basis for allocating target vehicle parking spaces and managing parking lot resources, and realizing the intelligent management of the parking lot.

[0033] Step S400: inputting the target vehicle image into a target detection model to obtain target vehicle information;

[0034] Specifically, the target detection model is a detection algorithm based on deep learning, such as Faster R-CNN, YOLOv3, etc. Through a large number of sample data, the detection model can obtain the ability to detect vehicles. The target detection model can accurately detect the position and range of the target vehicle in the target vehicle image, and output the attribute information of the target vehicle, such as vehicle type, license plate number, color, etc. First, the target vehicle image samples containing various types of vehicles and the corresponding annotation information need to be extracted from the data set as the training data of the target detection model; then the network structure of the target detection model is constructed using the deep learning framework, including convolutional layer, pooling layer and fully connected layer; then the training data is input into the target detection model, and the parameters of the model are adjusted through the back propagation algorithm, so that the model learns to detect the target vehicle from the input image and analyze its attribute information. The trained target detection model can be used to detect new input target vehicle images and accurately output the position, vehicle type, license plate number, etc. of the target vehicle.

[0035] The acquired target vehicle image is input into the target detection model to analyze the specific information of the target vehicle, which provides a basis for identifying the identity of the target vehicle and managing its parking. Through the target detection technology, irrelevant information in the input image is filtered, and the relevant attributes of the target vehicle are accurately obtained, realizing automatic management.

[0036] Step S500: traversing the vehicle database to obtain first vehicle information and matching with the target vehicle information;

[0037] Specifically, the vehicle database stores the relevant information of all target vehicles that have entered the field, such as license plate number, vehicle type, and regular parking position, as a reference for identifying the identity of the target vehicle. First, the vehicle database fields corresponding to the target vehicle information are constructed, including license plate number, vehicle type, and vehicle owner information; then SQL language is used to traverse the vehicle database to obtain first vehicle information; then the first vehicle information is matched with the target vehicle information one by one to determine whether there is a corresponding relationship in terms of license plate number, vehicle type, etc. If the target vehicle information has a high matching degree with a certain first vehicle information, it means that the identity of the target vehicle corresponds to the first vehicle, and the historical parking information of the target vehicle is obtained accordingly to allocate a regular parking space.

[0038] By querying the target vehicle information in the vehicle database, the identity of the target vehicle is automatically identified without human intervention. The vehicle information is stored in the database and the information matching is performed by the query algorithm, which efficiently manages a large amount of vehicle data and provides convenience for identifying the identity of the target vehicle and obtaining historical information.

[0039] Step S600: When the target vehicle information matches the first vehicle information, obtaining the target vehicle historical parking information according to the first vehicle information, and assigning a parking space for the target vehicle according to the target vehicle historical parking information and the parking distribution information;

[0040] Specifically, after confirming the identity of the target vehicle by matching the first vehicle information, the historical parking information of the target vehicle is obtained according to the first vehicle information, and then a parking space is assigned for the target vehicle according to the historical parking information and the obtained parking distribution information. The first vehicle information records various information corresponding to the target vehicle, including historical parking information such as regular parking space, parking duration, etc., as a reference for recommending a parking space for the target vehicle.

[0041] A parking space recommendation model is constructed, wherein the model input is the historical parking information of the target vehicle, such as regular parking space, parking duration, etc., and the vehicle distribution information parsed from the panoramic image of the parking lot, including the occupancy state of each parking space. Then, the parking preference features of the target vehicle are extracted from the historical parking information, such as the preference for parking spaces near the entrance or elevator; the features of the current available parking spaces are extracted from the parking distribution information, such as location, surrounding environment, etc. Secondly, according to the extracted features, some parking spaces that do not meet the requirements are filtered out, such as parking spaces far from the entrance. The filtered parking space set is the candidate parking space. Then, according to different features, a scoring mechanism is set for the candidate parking spaces, such as adding 5 points for being close to the regular parking space, and adding 3 points for having no other cars in the surrounding area. The higher the score, the better the parking space. Then, the candidate parking spaces with similar locations are clustered, and each cluster represents a recommended area to provide the target vehicle with options. Then, based on the parking space scoring and candidate parking space clustering, the top N parking spaces or M parking spaces in each cluster are selected and recommended to the target vehicle as parking spaces.

[0042] The target vehicle is intelligently selected a suitable parking space by the parking space recommendation model, which not only meets the parking preference of the target vehicle, but also meets the current vehicle distribution situation of the parking lot, realizes the rational allocation of resources, improves the space utilization rate, reduces the artificial burden, and improves the allocation efficiency.

[0043] Step S700: When the target vehicle information does not match the first vehicle information, adding the target vehicle to the vehicle database, interacting with the target vehicle, obtaining the destination of the target vehicle, and assigning a parking space for the target vehicle according to the destination and the parking distribution information;

[0044] Specifically, when the target vehicle information fails to match the first vehicle information and the identity of the target vehicle cannot be identified in the vehicle database, the target vehicle information is added to the vehicle database, then the destination of the target vehicle is obtained through human-computer interaction, and finally a parking space is assigned for the target vehicle according to the destination and the parking distribution information.

[0045] First, the acquired target vehicle information is added as a new entry to insert the vehicle database, and the added information includes license plate number, vehicle model, entry time, etc., which will be recorded for future identification of the target vehicle. Then, a human-computer interaction function is configured for communication with the driver of the target vehicle to obtain the destination information of the target vehicle this time, and the human-computer interaction is performed through a display screen, voice, etc. Let the driver select the destination category, such as office, shopping, dining, etc., or input the specific store name to provide more detailed information. Finally, according to the obtained destination information and the current parking distribution information, a parking space is recommended for the target vehicle. For example, if the destination is a certain mall, according to the mall entrance location and the parking lot empty space, the empty space near the entrance is recommended; if the specific store is provided, the parking space near the store can be recommended.

[0046] By adding new vehicle information to the database and human-computer interaction, the identity information and destination of the newly entered target vehicle are obtained, and then a parking space is allocated according to these information, realizing the automatic parking service of new vehicles and providing support for improving the parking convenience of new vehicles.

[0047] Step S800: obtaining the off-site information of the target vehicle through the off-site camera, and adding the off-site information to the vehicle database.

[0048] Specifically, when the target vehicle exits the parking lot, the off-site information of the target vehicle is obtained through the off-site camera, and the information is added to the vehicle database to complete the recording of the target vehicle's parking information this time.

[0049] When the off-site camera monitors the target vehicle exiting the parking lot, it will automatically identify and track it to obtain video stream information containing the off-site of the target vehicle. Then the video stream needs to be analyzed to extract the off-site information of the target vehicle, such as off-site time and exit location. First, a target detection algorithm is used to detect the video stream to find the image frame containing the target vehicle, and then a target recognition model is used to identify the license plate number of the vehicle to determine its identity. The timestamp of the extracted video stream is the off-site time of the target vehicle. After obtaining the off-site information of the target vehicle, the information is added to the corresponding vehicle entry in the vehicle database to update the information data of the target vehicle in the database.

[0050] By obtaining the off-site information of the target vehicle through off-site monitoring and updating it to the vehicle database, the information management of the vehicle is improved, the information is automatically obtained and added, the burden of manual recording is reduced, and the accuracy of the information is improved.

[0051] Further, as shown in Figure 2 the embodiments of the present application further include:

[0052] Step S210: When the entrance camera detects the target vehicle, the target vehicle is tracked and photographed to obtain an entrance video stream;

[0053] Step S220: The entrance video stream is subjected to image frame screening to obtain a vehicle image set, all vehicle images in the target vehicle image set containing the target vehicle;

[0054] Step S230: The vehicle image set is traversed, and the vehicle image is subjected to quality evaluation by an image quality evaluation model to obtain a target vehicle image.

[0055] Specifically, first, when the entrance camera detects that the target vehicle enters the monitoring range, it is tracked and photographed to obtain an entrance video stream containing the driving process of the target vehicle. The entrance video stream records the entire process of the target vehicle from entering the entrance to driving out of the monitoring range. Then, the entrance video stream is processed, the moving target in the video stream is detected by image processing techniques such as frame difference, and the image frame containing the target vehicle is extracted to obtain a set of vehicle images. This set of images contains different angle images of the target vehicle.

[0056] Next, a vehicle image is taken from the vehicle image set, and a plurality of features of the image are obtained using a feature extraction algorithm, such as sharpness features, contrast features, etc. The features are input into an image quality evaluation model to obtain corresponding scores, such as an SSIM score of 0.8, a contrast score of 0.7, etc. The overall score is calculated according to the set weight, such as 0.8*0.6+0.7*0.3=0.74. The image with the highest score is selected as the target vehicle image according to the score result.

[0057] By accurately extracting the best image of the target vehicle from the entrance video stream, an accurate target vehicle image is provided for subsequent vehicle detection and identification, filtering out useless information to support vehicle monitoring and management.

[0058] Further, as shown in Figure 3 the embodiments of the present application also include:

[0059] Step S310: The collected panoramic video stream is preprocessed by denoising, brightness adjustment, and contrast enhancement to obtain a set of images to be used;

[0060] Step S320: The set of images to be used is subjected to image extraction, and all images to be used are spliced to obtain a parking lot panoramic image;

[0061] Step S330: A parking distribution recognition model is constructed, and the parking lot panoramic image is input into the parking distribution model to obtain parking distribution information.

[0062] Specifically, first, the panoramic video stream collected by the panoramic camera may have noise, uneven lighting and other problems, affecting the image quality. Therefore, the video stream is preprocessed to improve the image quality through denoising, brightness adjustment, contrast enhancement and other techniques to obtain a set of clear images. Then, images are extracted from the image set, and the extracted images are spliced to connect the registered pictures into a large picture to obtain a panoramic image, which clearly shows the distribution information of vehicles and objects in the parking lot. Finally, the obtained panoramic image of the parking lot is input into the parking distribution recognition model to analyze the vehicle distribution information in the image, detect the position, posture and occupied area of each vehicle on the panoramic image, and obtain information such as the number of vehicles, whether each parking space has a vehicle, and the gap between parking spaces.

[0063] Through the transformation from the video stream to the parking distribution information, the image preprocessing, splicing and recognition model are used to realize the automatic extraction of information, provide support for obtaining accurate vehicle distribution information, and improve the parking lot monitoring efficiency.

[0064] Further, the embodiments of the present application also include:

[0065] Step S331: acquiring a parking lot image data set by using big data technology;

[0066] Step S332: filtering the parking lot image data set to obtain a sample image data set;

[0067] Step S333: traversing the sample image data set, and performing vehicle parking annotation on the sample image, the parking annotation including idle and occupied, wherein each sample image corresponds to a sample annotation set;

[0068] Step S334: traversing the sample annotation set, and segmenting the sample image corresponding to the sample annotation set to obtain a sample data set;

[0069] Step S335: dividing the sample data set and the sample annotation set into a sample training set and a sample verification set;

[0070] Step S336: performing supervised learning training on the convolutional neural network model through the sample training set, verifying the parking distribution recognition model through the sample verification set, and obtaining the parking distribution recognition model.

[0071] Specifically, first, a large number of parking lot panoramic images are obtained from various sources using big data technology to form a parking lot image dataset, realizing the collection, cleaning and integration of massive data. Second, the obtained image dataset is filtered based on image clarity, resolution and other standards, and low-quality images are filtered out to obtain a high-quality sample image dataset. Then, the sample image dataset is traversed, and each sample image is labeled for vehicle parking, which is divided into two categories: free parking spaces and occupied parking spaces. Each sample image corresponds to a sample label set. After that, the sample label set is traversed, and the corresponding sample image is segmented according to the label information to obtain a sample dataset containing parking area and vehicle area. Next, the obtained sample dataset and sample label set are divided into training set and validation set in proportion, and the training set is used for model training and the validation set is used for model evaluation. Finally, the deep convolutional neural network model is trained through supervised learning training set, and the model obtained by training is verified and evaluated through validation set, and the final parking distribution recognition model is obtained.

[0072] By constructing the dataset, labeling and training the deep learning model, the parking distribution recognition model is obtained, the accurate mapping of panoramic images and vehicle distribution information is realized, and the full-automatic vehicle state monitoring is realized.

[0073] Further, the embodiments of the application also include:

[0074] Step S610: data analysis is performed on the target historical parking information to obtain a historical parking duration set;

[0075] Step S620: the historical parking duration set is traversed, and the historical parking duration is sorted in descending order of duration;

[0076] Step S630: set a parking duration threshold, divide the sorted historical parking duration according to the parking duration threshold, and take the historical parking data with a historical parking duration greater than or equal to the parking duration threshold as the first reference data, and take the historical parking data with a historical parking duration less than the parking duration threshold as the second reference data;

[0077] Step S640: filtering the target vehicle historical parking information according to the first reference data to obtain a first recommended parking space set;

[0078] Step S650: filtering the target vehicle historical parking information according to the second reference data to obtain a second recommended parking space set;

[0079] Step S660: set a parking space recommendation threshold, match the first recommended parking space set with the parking distribution information to obtain a target recommended parking space set;

[0080] Step S670: When the number of parking spaces in the target recommended parking space set is greater than or equal to the parking space recommendation threshold, the target recommended parking space set is sent to the target vehicle, and a recommendation index is identified.

[0081] Specifically, first, the historical parking information of the target vehicle is analyzed to obtain a historical parking duration set, recording the specific duration of each parking. Second, the historical parking duration set is sorted in descending order to obtain the distribution of the target vehicle parking duration. Then, a parking duration threshold is set, and the sorted historical parking duration set is divided according to the threshold. The duration greater than the threshold is classified as the first reference data, and the duration less than the threshold is classified as the second reference data. The first reference data can recommend long-time parking spaces, and the second reference data can recommend short-time parking spaces. Then, the historical parking information is filtered according to the first reference data to obtain a first recommended parking space set; the second recommended parking space set is obtained according to the second reference data, and the recommended parking space corresponds to the position frequently parked by the target vehicle in the past. Next, a parking space recommendation threshold is set, and the target recommended parking space set is obtained by matching the first recommended parking space set with the current parking distribution information. The target recommended parking space set is the recommended result considering the current parking vacancy. Finally, when the number of parking spaces in the target recommended parking space set is greater than or equal to the recommendation threshold, the set is sent to the target vehicle, and a recommendation index is identified, completing the parking space recommendation.

[0082] By analyzing historical data and matching current distribution, personalized parking space recommendation is provided for the target vehicle, guiding the vehicle to select the parking position of habit, improving driving convenience, and realizing traffic guidance, avoiding random parking of vehicles, which is beneficial to the rational use and management of parking space.

[0083] Further, the embodiments of the present application also include:

[0084] Step S681: When the number of parking spaces in the target recommended parking space set is less than the parking space recommendation threshold, the number of remaining recommended parking spaces is obtained according to the target recommended parking space set and the parking space recommendation threshold.

[0085] Step S682: The second recommended parking space set is matched with the parking distribution information to obtain the recommendation index of each parking space in the second recommended parking space set.

[0086] Step S683: N parking spaces are extracted from the second recommended parking space set according to the recommendation index, where N is the number of remaining recommended parking spaces.

[0087] Specifically, when the number of parking spaces in the target recommended parking space set is less than the preset parking space recommendation threshold, the number of remaining recommended parking spaces is calculated according to the target recommended parking space set and the threshold, and the number of remaining recommended parking spaces is the number of parking spaces that need to be increased to reach the recommendation threshold. First, the second recommended parking space set is matched with the current parking distribution information, and a recommendation index is calculated for each parking space in the second recommended parking space set to evaluate the idle possibility and matching degree of each parking space. Then, the parking spaces with the number of remaining recommended parking spaces are extracted from the second recommended parking space set according to the recommendation index, and the extracted parking spaces have a higher recommendation index and a greater idle possibility, serving as a supplement to the recommendation.

[0088] By calculating the number of remaining recommended parking spaces on the basis of the original recommended scheme, matching the second recommended parking space set with the parking distribution information, and extracting the number of required parking spaces according to the scoring result, the target recommended parking space set is supplemented, the parking demand of the vehicle is maximally met, and the selection freedom and parking convenience are improved.

[0089] Further, the embodiments of the present application also include:

[0090] Step S710: When the target vehicle information does not exist in the vehicle database, establishing the target vehicle database field according to the target vehicle information;

[0091] Step S720: According to the destination of the target vehicle, traversing the vehicle database to obtain a target parking space to be selected;

[0092] Step S730: Based on the matching between the target parking space to be selected and the parking distribution information, selecting a space parking space as a target recommended parking space set and sending it to the target vehicle.

[0093] Specifically, when the target vehicle information does not exist in the existing vehicle database, the database field is established according to the information provided by the target vehicle, and the information file is created to record the parking information and habits of the vehicle, thereby providing support for the recommendation. First, according to the destination of the target vehicle this time, the commonly used parking space of the user going to the destination is found in the vehicle database as a target parking space to be selected, thereby meeting the parking demand this time. Then, the target parking space is matched with the current parking distribution information based on the selection, and an idle parking space is selected as a target recommended parking space set and sent to the target vehicle.

[0094] By creating an information file, querying a database, and matching a distribution, the parking space recommendation for a new vehicle is realized, a new vehicle parking space recommendation scheme is realized, a personalized recommendation service is quickly provided for a stranger vehicle, and the technical effects of full monitoring of a parking lot video and intelligent parking space allocation of a vehicle are achieved.

[0095] In summary, the high-position video monitoring method for parking lots based on deep recognition provided by the embodiment has the following technical effects:

[0096] The high-position camera group is constructed, and the high-position camera group includes an entrance camera, a panoramic camera, and an exit camera, which are used to acquire vehicle and parking lot video images; the target vehicle is photographed by the entrance camera to acquire a target vehicle image, thereby providing an image basis for acquiring vehicle information; the parking lot is photographed by the panoramic camera to acquire a panoramic image of the parking lot, and parking distribution information is acquired according to the panoramic image of the parking lot, thereby providing a basis for intelligent allocation of parking spaces for vehicles; the target vehicle image is input into a target detection model to acquire target vehicle information, and vehicle information is acquired according to the vehicle image, thereby providing information support for parking space recommendation; the vehicle database is traversed to acquire first vehicle information, and the first vehicle information is matched with the target vehicle information; when the target vehicle information matches the first vehicle information, target vehicle historical parking information is acquired according to the first vehicle information, and a parking space is allocated for the target vehicle according to the target vehicle historical parking information and the parking distribution information; if the vehicle database contains the vehicle information, a parking space is intelligently selected for the vehicle according to the historical parking information and real-time parking space information; when the target vehicle information does not match the first vehicle information, the target vehicle is added to the vehicle database, and the target vehicle is interacted to acquire a destination of the target vehicle, and a parking space is allocated for the target vehicle according to the destination and the parking distribution information; if the vehicle database does not contain the vehicle information, the destination information of the vehicle is acquired by interaction with the vehicle owner, and a parking space is allocated for the vehicle according to the destination information and real-time parking space information; the exit information of the target vehicle is acquired by the exit camera, and the exit is added to the vehicle database, and the information of the vehicle in the vehicle database is updated, thereby achieving the technical effects of parking lot video monitoring and intelligent parking space allocation for vehicles.

[0097] Embodiment two

[0098] Based on the same inventive concept as the high-position video monitoring method for parking lots based on deep recognition in the foregoing embodiments, as shown in Figure 4 The embodiment provides a high-position video monitoring system for parking lots based on deep recognition, and the system comprises:

[0099] The camera group construction module 11 is configured to construct a high-position camera group, and the high-position camera group includes an entrance camera, a panoramic camera, and an exit camera;

[0100] The target vehicle image module 12 is configured to photograph a target vehicle by the entrance camera to acquire a target vehicle image;

[0101] The parking distribution information module 13 is configured to photograph a parking lot by the panoramic camera to acquire a panoramic image of the parking lot, and acquire parking distribution information according to the panoramic image of the parking lot;

[0102] a target vehicle information module 14, configured to input the target vehicle image into a target detection model to obtain target vehicle information;

[0103] a vehicle information matching module 15, configured to traverse a vehicle database to obtain first vehicle information and match the first vehicle information with the target vehicle information;

[0104] a vehicle information existence module 16, configured to, when the target vehicle information matches the first vehicle information, obtain target vehicle historical parking information according to the first vehicle information, and allocate a parking space for the target vehicle according to the target vehicle historical parking information and the parking distribution information;

[0105] a vehicle information absence module 17, configured to, when the target vehicle information does not match the first vehicle information, add the target vehicle to the vehicle database, interact with the target vehicle, obtain a destination of the target vehicle, and allocate a parking space for the target vehicle according to the destination and the parking distribution information;

[0106] a vehicle off-site information module 18, configured to obtain off-site information of the target vehicle through the off-site camera and add the off-site information to the vehicle database.

[0107] Further, the target vehicle image module 12 includes the following execution steps:

[0108] when the target vehicle is detected by the entry camera, tracking and photographing the target vehicle to obtain an entry video stream;

[0109] performing image frame screening on the entry video stream to obtain a vehicle image set, all vehicle images in the target vehicle image set containing the target vehicle;

[0110] traversing the vehicle image set, performing quality evaluation on the vehicle images through an image quality evaluation model to obtain a target vehicle image.

[0111] Further, the parking distribution information module 13 includes the following execution steps:

[0112] performing preprocessing on the collected panoramic video stream through denoising, brightness adjustment, and contrast enhancement to obtain a standby image set;

[0113] performing image extraction on the standby image set, and splicing all standby images to obtain a parking lot panoramic image;

[0114] constructing a parking distribution recognition model, inputting the parking lot panoramic image into the parking distribution model to obtain parking distribution information.

[0115] Further, the parking distribution information module 13 further includes the following execution steps:

[0116] acquiring a parking lot image dataset by using big data technology;

[0117] screening the parking lot image dataset to obtain a sample image dataset;

[0118] traversing the sample image dataset to perform vehicle parking labeling on sample images, the parking labeling including idle and occupied, wherein each sample image corresponds to a sample labeling set;

[0119] traversing the sample labeling set to segment the sample images corresponding to the sample labeling set to obtain a sample dataset;

[0120] dividing the sample dataset and the sample labeling set into a sample training set and a sample verification set;

[0121] performing supervised learning training on a convolutional neural network model through the sample training set, verifying the parking distribution identification model through the sample verification set, and obtaining a parking distribution identification model.

[0122] Further, the vehicle information existence module 16 includes the following execution steps:

[0123] performing data analysis on the target historical parking information to obtain a historical parking duration set;

[0124] traversing the historical parking duration set to sort the historical parking durations in descending order of duration;

[0125] setting a parking duration threshold, dividing the sorted historical parking durations according to the parking duration threshold, taking historical parking data with a historical parking duration greater than or equal to the parking duration threshold as first reference data, and taking historical parking data with a historical parking duration less than the parking duration threshold as second reference data;

[0126] screening the target vehicle historical parking information according to the first reference data to obtain a first recommended parking space set;

[0127] screening the target vehicle historical parking information according to the second reference data to obtain a second recommended parking space set;

[0128] setting a parking space recommendation threshold, matching the first recommended parking space set with the parking distribution information to obtain a target recommended parking space set;

[0129] when the number of parking spaces in the target recommended parking space set is greater than or equal to the parking space recommendation threshold, sending the target recommended parking space set to the target vehicle and identifying a recommendation index.

[0130] Further, the vehicle information existence module 16 further comprises the following execution steps:

[0131] When the number of parking spaces in the target recommended parking space set is less than the parking space recommendation threshold, the number of remaining recommended parking spaces is obtained according to the target recommended parking space set and the parking space recommendation threshold;

[0132] The second recommended parking space set is matched with the parking distribution information to obtain a recommendation index of each parking space in the second recommended parking space set;

[0133] N parking spaces are extracted from the second recommended parking space set according to the recommendation index, where N is the number of remaining recommended parking spaces.

[0134] Further, the vehicle information absence module 17 comprises the following execution steps:

[0135] When the target vehicle information does not exist in the vehicle database, the target vehicle database field is established according to the target vehicle information;

[0136] According to the destination of the target vehicle, the vehicle database is traversed to obtain a candidate target parking space;

[0137] The candidate target parking space is matched with the parking distribution information, and a space parking space is selected as a target recommended parking space set and sent to the target vehicle.

[0138] Any step of the above method can be stored as computer instructions or programs in an unlimited computer memory and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application. No redundant limitations are made here.

[0139] Further, the above-mentioned first or second may not only represent an order relationship, but also may represent a specific concept, and / or refer to the selection of individual or all elements. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for high-position video monitoring of parking lots based on depth recognition, the method being applied to a high-position video monitoring system for parking lots, characterized in that, The method includes: Construct a high-position camera group, which includes an entrance camera, a panoramic camera, and an exit camera; The entrance camera captures images of the target vehicle. The panoramic camera captures a panoramic image of the parking lot, and parking distribution information is obtained based on the panoramic image. The target vehicle image is input into the target detection model to obtain target vehicle information; Traverse the vehicle database, obtain the first vehicle information, and match it with the target vehicle information; When the target vehicle information matches the first vehicle information, the target vehicle’s historical parking information is obtained based on the first vehicle information, and a parking space is allocated to the target vehicle based on the target vehicle’s historical parking information and the parking distribution information. When the target vehicle information does not match the first vehicle information, the target vehicle is added to the vehicle database, the target vehicle is interacted with, the destination of the target vehicle is obtained, and a parking space is allocated to the target vehicle according to the destination and the parking distribution information. The departure information of the target vehicle is acquired through the exit camera, and the departure information is added to the vehicle database. The step of allocating parking spaces for the target vehicle based on the target vehicle's historical parking information and the parking distribution information includes: Data analysis is performed on the target's historical parking information to obtain a set of historical parking durations; Traverse the set of historical parking durations and sort the historical parking durations in descending order of duration; Set a parking duration threshold, divide the sorted historical parking durations according to the parking duration threshold, use historical parking data with historical parking duration greater than or equal to the parking duration threshold as first reference data, and use historical parking data with historical parking duration less than the parking duration threshold as second reference data; Based on the first reference data, the historical parking information of the target vehicle is filtered to obtain a first recommended parking space set; Based on the second reference data, the historical parking information of the target vehicle is filtered to obtain a second recommended parking space set; Set a parking space recommendation threshold, and match the first recommended parking space set with the parking distribution information to obtain a target recommended parking space set; When the number of parking spaces in the target recommended parking space set is greater than or equal to the parking space recommendation threshold, the target recommended parking space set is sent to the target vehicle, and a recommendation index is assigned. When the number of parking spaces in the target recommended parking space set is less than the parking space recommendation threshold, the remaining number of recommended parking spaces is obtained based on the target recommended parking space set and the parking space recommendation threshold. The recommendation index of each parking space in the second recommended parking space set is obtained by matching the parking distribution information with the second recommended parking space set. N parking spaces are extracted from the second recommended parking space set according to the recommendation index, where N is the number of remaining recommended parking spaces.

2. The method as described in claim 1, characterized in that, The step of capturing images of the target vehicle using the entrance camera to obtain the target vehicle image includes: When the entrance camera detects the target vehicle, it tracks and captures the target vehicle to obtain the entrance video stream; The incoming video stream is filtered by image frames to obtain a vehicle image set, and all vehicle images in the target vehicle image set contain the target vehicle. The vehicle image set is traversed, and the vehicle images are evaluated using an image quality assessment model to obtain the target vehicle image.

3. The method as described in claim 1, characterized in that, The step of capturing a panoramic image of the parking lot using the panoramic camera and obtaining parking distribution information based on the panoramic image includes: The acquired panoramic video stream is preprocessed by denoising, brightness adjustment, and contrast enhancement to obtain a set of images to be used. Image extraction is performed on the set of images to be used, and all images to be used are stitched together to obtain a panoramic image of the parking lot; A parking distribution recognition model is constructed by inputting the panoramic image of the parking lot into the parking distribution recognition model to obtain parking distribution information.

4. The method as described in claim 3, characterized in that, The construction of the parking distribution identification model includes: Utilizing big data technology to obtain parking lot image datasets; The parking lot image dataset is filtered to obtain a sample image dataset; Traverse the sample image dataset and annotate the sample images with vehicle parking information. The parking information includes free and occupied. Each sample image corresponds to a sample annotation set. Traverse the sample annotation set, segment the sample images corresponding to the sample annotation set, and obtain the sample dataset; The sample dataset and the sample annotation set are divided into a sample training set and a sample validation set; The convolutional neural network model is trained using the sample training set under supervised learning conditions, and the parking distribution recognition model is validated using the sample validation set to obtain the parking distribution recognition model.

5. The method as described in claim 1, characterized in that, When the target vehicle information does not match the first vehicle information, the target vehicle is added to the vehicle database, the target vehicle is interacted with to obtain its destination, and a parking space is allocated to the target vehicle based on the destination and the parking distribution information, including: If the target vehicle information does not exist in the vehicle database, the target vehicle database field is created based on the target vehicle information; Based on the destination of the target vehicle, traverse the vehicle database to obtain candidate target parking spaces; Based on the matching of the candidate target parking spaces with the parking distribution information, the selected parking spaces are used as the target recommended parking space set and sent to the target vehicle.

6. A high-position video monitoring system for parking lots based on depth recognition, characterized in that, The system includes: A camera group construction module is used to construct a high-position camera group, which includes an entrance camera, a panoramic camera, and an exit camera. The target vehicle image module is used to capture images of the target vehicle through the entrance camera. The parking distribution information module is used to capture a panoramic image of the parking lot using the panoramic camera, and to obtain parking distribution information based on the panoramic image of the parking lot. The target vehicle information module is used to input the target vehicle image into the target detection model to obtain target vehicle information. The vehicle information matching module is used to traverse the vehicle database, obtain first vehicle information, and match it with target vehicle information. The vehicle information storage module is used to, when the target vehicle information matches the first vehicle information, obtain the target vehicle's historical parking information based on the first vehicle information, and allocate parking spaces to the target vehicle based on the target vehicle's historical parking information and the parking distribution information. The vehicle information missing module is used to add the target vehicle to the vehicle database when the target vehicle information does not match the first vehicle information, interact with the target vehicle, obtain the destination of the target vehicle, and allocate a parking space for the target vehicle according to the destination and the parking distribution information. A vehicle departure information module is used to acquire departure information of the target vehicle through the departure camera and add the departure information to the vehicle database. The vehicle information missing module is also used to allocate parking spaces to the target vehicle based on the target vehicle's historical parking information and the parking distribution information, including: Data analysis is performed on the target's historical parking information to obtain a set of historical parking durations; Traverse the set of historical parking durations and sort the historical parking durations in descending order of duration; Set a parking duration threshold, divide the sorted historical parking durations according to the parking duration threshold, use historical parking data with historical parking duration greater than or equal to the parking duration threshold as first reference data, and use historical parking data with historical parking duration less than the parking duration threshold as second reference data; Based on the first reference data, the historical parking information of the target vehicle is filtered to obtain a first recommended parking space set; Based on the second reference data, the historical parking information of the target vehicle is filtered to obtain a second recommended parking space set; Set a parking space recommendation threshold, and match the first recommended parking space set with the parking distribution information to obtain a target recommended parking space set; When the number of parking spaces in the target recommended parking space set is greater than or equal to the parking space recommendation threshold, the target recommended parking space set is sent to the target vehicle, and a recommendation index is assigned. When the number of parking spaces in the target recommended parking space set is less than the parking space recommendation threshold, the remaining number of recommended parking spaces is obtained based on the target recommended parking space set and the parking space recommendation threshold. The recommendation index of each parking space in the second recommended parking space set is obtained by matching the parking distribution information with the second recommended parking space set. N parking spaces are extracted from the second recommended parking space set according to the recommendation index, where N is the number of remaining recommended parking spaces.

Images