A method and system for intelligent survey of urban parking resources
Through intelligent video recognition and drone aerial photography technology combined with image processing algorithms, an intelligent census of urban parking resources was carried out, which solved the problem of unreasonable distribution of parking resources, achieved rapid and accurate acquisition of berth information, and supported reasonable parking facility planning.
Patent Information
- Application Number
- CN202410141752.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-01
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-02-01
AI Technical Summary
The existing technology cannot effectively conduct a rapid and intelligent census of large-scale parking resources in urban areas, resulting in the unreasonable distribution of parking resources and the shortage of parking resources being difficult to solve.
Using a method based on intelligent video recognition and drone aerial photography, combined with image classification recognition, object detection, OCR text recognition and edge detection algorithms, information acquisition and statistics are obtained and counted on public berths and building berths on the road and off-road, and berths are established. Location map database and ensure the accuracy of data through regular audit mechanisms.
It greatly saves labor costs, improves the efficiency of parking resource census, ensures the accuracy and authenticity of berth information, and supports municipal departments to carry out reasonable parking facilities planning.
Smart Images

Figure CN117975761B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of parking facility information management, and in particular to an intelligent survey method and system for urban parking resources. Background Art
[0002] With the development of cities and social progress, the rapid increase in car maintenance has made the problem of unreasonable distribution of parking resources in urban areas and the problem of parking resource shortage increasingly prominent, resulting in frequent parking difficulties and illegal parking in densely populated urban areas;
[0003] To solve this problem, the municipal government needs to make further reasonable planning for parking facilities. Before reasonable planning, a comprehensive census of parking resources in the urban area is needed to provide data support for subsequent reasonable planning. The existing parking space census methods mostly rely on manual surveys, which have high manpower and time investment costs. In the context of big data, large-scale urban video data based on vehicle-mounted and drone aerial photography and deep learning algorithms have made large-scale parking space surveys possible. To this end, we provide a method and system for intelligent census of urban parking resources. Summary of the invention
[0004] The purpose of the present invention is to provide a method and system for intelligent survey of urban parking resources.
[0005] The technical problem solved by the present invention is: how to conduct a comprehensive, rapid and intelligent survey of parking resources in a large area of an urban area;
[0006] The present invention can be implemented by the following technical solutions: an intelligent census method for urban parking resources, including census of on-street public parking spaces, off-street public parking spaces and building accessory parking spaces;
[0007] For public parking spaces on the road, the system can intelligently identify the parking fee sign or parking space based on the video and extract the parking information, including the parking location, parking type, parking quantity and parking fee charging method;
[0008] For off-street public parking spaces, first import the parking space information through the parking lot registration and gate system. If the information cannot be imported, the parking space information is obtained by intelligently identifying the toll sign or parking space based on the video.
[0009] For parking spaces built with buildings, we use web crawlers to search the pre-approval public information online, extract the motor vehicle parking space construction information to determine the parking space information for the newly built community; and use video intelligent identification to count the number of parking spaces in open old communities.
[0010] A further technical improvement of the present invention is that for on-street public berths, video-based berth position recognition is mainly achieved by building a berth position map database to train an image classification recognition algorithm, then identifying the specific berth position, and further determining the berth type based on the berth position.
[0011] A further technical improvement of the present invention is that for on-street public berths and off-street public berths, the step of identifying the berth charging status and charging method comprises:
[0012] Extract berth image data frame by frame and use target detection algorithm to build berth fee sign recognition model to identify berth fee sign;
[0013] When no fee sign is identified, the corresponding berth is a free berth;
[0014] When a toll sign is identified and the result is "free", the corresponding berth is a free berth;
[0015] When there is a fee sign and the recognition result is "fee", the OCR text recognition algorithm is used to further extract the berth fee standard.
[0016] A further technical improvement of the present invention is that when there is a toll sign, the number of parking spaces can be directly extracted using the OCR text recognition algorithm; when there is no toll sign or the number of parking spaces is not marked on the toll sign, the Canny edge detector in open CV is used to perform contour detection on the preprocessed image to identify the parking space contour, and then identify the parking spaces and count the number of parking spaces.
[0017] A further technical improvement of the present invention is that the number of parking spaces in old residential areas can be determined by using the Canny edge detector in open CV to perform contour detection on the preprocessed image, identify the parking space contours, and then identify the parking spaces and count the number of parking spaces.
[0018] A further technical improvement of the present invention is that it also includes a regular review mechanism for berth information, and reviews the berth census data through regular on-site berth surveys and data comparisons. The regular review forms include static data error correction and dynamic data comparison.
[0019] A further technical improvement of the present invention is that: static data error correction includes error correction of berth locations, berth types and charging methods in berth census data to check whether there are duplicate or contradictory data;
[0020] Dynamic data comparison includes correcting the dynamic berth quantity data in the berth census data. By comparing the existing berth census data and the historical berth census data at the same location, the berth occupancy and usage frequency are compared to check whether the berth data changes are too large.
[0021] A further technical improvement of the present invention is that for old communities where information is missing, the community property management department can obtain berth information by scanning a QR code and filling out a questionnaire.
[0022] A system includes a communication unit for acquiring drone aerial photography and vehicle-mounted videos, a database for storing image data, and a processor for processing and identifying images.
[0023] Compared with the prior art, the present invention has the following beneficial effects:
[0024] (1) The present invention divides parking resources and obtains parking information for different parking spaces. Based on aerial video technology and vehicle-mounted video technology, the present invention conducts a census of parking space locations, parking space types, charging methods, and parking space quantities in a wide range of urban areas. By combining image classification and recognition algorithms, target detection algorithms, OCR text recognition algorithms, and edge detection algorithms, the present invention accurately determines the values of various parameters in the parking space information, thereby greatly saving labor costs and improving census efficiency.
[0025] (2) The present invention also ensures the authenticity and accuracy of the data by establishing a regular audit mechanism to review and correct the data. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0027] Figure 1 It is a logic flow chart of the survey of the present invention;
[0028] Figure 2 This is a diagram of the CTPN network architecture in the present invention;
[0029] Figure 3 This is the CRNN network architecture diagram in the present invention. DETAILED DESCRIPTION
[0030] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.
[0031] See also Figure 1-3 As shown, a method for intelligent survey of urban parking resources includes surveying three types of parking resources, namely, intelligent survey of on-street public parking spaces, intelligent survey of off-street public parking spaces, and intelligent survey of parking spaces provided with buildings. The intelligent survey methods for three different parking space areas are as follows;
[0032] A: Regarding the intelligent survey of public parking spaces on the road: intelligently identify parking spaces and extract parking information based on drone aerial videos or vehicle-mounted videos, where the parking information includes parking location, parking type, parking quantity and parking fee method; specifically,
[0033] A1: The identification of berth location includes the following steps:
[0034] (1) Based on drone aerial videos and vehicle-mounted videos, berth images at different locations are extracted to build a berth location map database;
[0035] (2) Use image classification and recognition algorithms such as convolutional neural network (CNN) and support vector machine (SVM) to train the berth position recognition model;
[0036] (3) Extract the parking space location image from the real-time collected video, and use the trained parking space location recognition model to identify the specific parking space location (motor vehicle lane, non-motor vehicle lane, sidewalk, etc.).
[0037] A2: Parking spaces can be divided into on-street parking spaces and roadside parking spaces. On-street parking spaces are located on one or both sides of the road, while roadside parking spaces are located on non-motorized vehicle lanes or sidewalks.
[0038] Based on the above-identified parking space positions, the parking space type is identified to distinguish between on-street parking spaces and roadside parking spaces; if the parking space is located on a motor vehicle lane, it is determined to be an on-street parking space; if the parking space is located on a non-motor vehicle lane or a sidewalk, it is determined to be a roadside parking space;
[0039] A3: Different berths have different charging methods. The steps to identify the berth charging status and method include:
[0040] A31: First, we extract the parking space image data frame by frame based on the video data, and use the target detection algorithm R-CNN to build a parking space fee sign recognition model to identify the parking space fee sign; specifically,
[0041] a) Input: The berth image data with and without the toll sign is used as the training data set;
[0042] b) Data conversion: converting the berth image data into a two-dimensional pixel matrix;
[0043] c) Candidate area selection: using a selective search algorithm to extract candidate areas that may contain charges from the berth image data from bottom to top;
[0044] d) Data normalization: Scale the candidate area image of the toll sign to a size of 227*227;
[0045] e) Training the parking fee sign recognition model, including convolution, pooling and full connection operations;
[0046] Convolution: Extract the features of the candidate area image through the convolution layer. Suppose the input image matrix size is w, the convolution kernel size is k, the stride is s, and the number of zero-filling layers is p. The size of the feature map w' generated after the convolution operation is completed is: w'=w+2p-k*s+1;
[0047] Pooling: Input the candidate region image into the pooling layer, compress the image features and reduce the dimension through the pooling layer. Suppose the sliding window of the pooling layer is w1, the stride is s1, and each time it slides to a region, select the maximum value or mean value of the region as the output of the pooling layer;
[0048] Full connection: The candidate region feature maps obtained through multi-layer convolution and pooling operations are connected into vectors by row and input into the fully connected layer;
[0049] Output Y = φ(x). φ() is the activation function, including sigmoid, Tanh, Relu functions, etc.
[0050] f) The berth image data extracted based on the video data is input into the trained berth fee sign recognition model, and the Softmax classifier is used to calculate the probability of the image containing a fee sign or not containing a fee sign, and the category with the highest probability is selected as the berth fee.
[0051] When the recognition result is no fee sign, the parking space is free; otherwise, if a fee sign is recognized, the Optical Character Recognition (OCR) text recognition algorithm can be used to identify whether there is a fee; if the fee sign recognition result is "free", the parking space is free; otherwise, if the fee sign recognition result is "paid", the parking space is paid;
[0052] A32: When the recognition result is "charge", the OCR text recognition algorithm is used to further extract the berth charge standard. The specific operation is as follows:
[0053] a) Input: image of the fee sign containing berth fee information;
[0054] b) Image preprocessing: Binarize the image and convert it into a grayscale image. At the same time, perform image noise reduction and tilt correction to improve image quality.
[0055] c) Text detection: The commonly used algorithm CTPN neural network (Connectionist Text Proposal Network) is used to detect the position of the relevant text on the berth fee sign. The specific detection steps are as follows:
[0056] ①Use VGG16 (Visual Geometry Group 16) network architecture to extract spatial features;
[0057] ② Use a 3×3 sliding window to reassemble the feature map;
[0058] ③ Input the reorganized feature map into the bidirectional LSTM to extract the sequence features of each row;
[0059] ④ Input the bi-LSTM output sequence into a convolutional layer (FC);
[0060] ⑤ Input the output of the previous convolution into the three prediction branches of 2k scores, 2k vertical coordinates and side-refinement, and mark the location of the berth fee sign with a rectangle;
[0061] ⑥ Based on the text line construction algorithm, the classified text rectangles are merged into text lines to realize text detection.
[0062] d) Text recognition: The text on the berth fee sign is recognized using a convolutional neural network (CRNN) structure. CRNN includes a convolutional layer, a recurrent layer, and a recognition layer:
[0063] Convolutional layer: Use CNN to extract feature sequences from the input image;
[0064] Recurrent layer: Use RNN to predict the distribution of feature sequence labels obtained from the convolutional layer;
[0065] Recognition layer: The label distribution obtained from the circulation layer using CTC is converted into the berth fee sign text recognition result through operations such as de-duplication and integration.
[0066] A4: The steps for extracting berth quantity are as follows:
[0067] For parking spaces with number on the toll sign, the parking space number information is directly extracted; for free parking spaces and toll signs without parking space number, pictures containing parking spaces are extracted from the video, denoised, grayscaled and binarized, and the Canny edge detector in open CV is used to perform contour detection on the image to identify the parking space contour and then identify the parking space.
[0068] Among them, the Canny edge detector in open CV is used to detect the contour of the berth image. The specific steps are as follows:
[0069] (1) Image preprocessing: Use Gaussian blurring to eliminate image noise;
[0070] (2) Calculate the brightness gradient value of the image: Use Sobel horizontal and vertical convolution kernels to filter the image, and calculate the amplitude G and direction θ of the brightness gradient. The specific calculation method is as follows:
[0071]
[0072] θ = tan -1 (G y / G x )
[0073] where Gx and Gy are the amplitudes of the brightness gradient in the X-axis direction and Y-axis direction respectively.
[0074] (3) Remove false edges: Use the Non-Maximum Suppression (NMS) algorithm to process the edges of the image;
[0075] (4) Threshold detection with hysteresis: Compare the gradient value G calculated in step (2) with the minimum threshold Gmin and the maximum threshold Gmax:
[0076] If G > Gmax, retain the pixel in the final edge map;
[0077] If G < Gmin, remove the pixel in the final edge map;
[0078] If Gmin ≤ G ≤ Gmax, mark the pixel as a weak edge. If the weak edge is connected to a strong edge, retain the pixel in the final edge map.
[0079] (5) Output the recognized berth contour image.
[0080] When the parking berth recognition is completed, count the total number of berths.
[0081] B: Regarding the intelligent census of off-street public berths, the berth information can be imported through the parking lot filing or the barrier gate system;
[0082] When the parking lot filing and the barrier gate system do not include the number of berths, extract the berth pictures from the UAV aerial video. First, identify the toll sign. For the toll sign with the number of berths, directly extract the berth number information;
[0083] For free berths and toll signs without the number of berths, extract the pictures containing parking berths from the video, perform grayscale processing on the pictures, use the Canny edge detector in open CV to identify the parking space markings, and then identify the parking berths. When the parking berth recognition is completed, count the total number of berths.
[0084] C: Regarding the intelligent census of building-matched berths, it is mainly divided into under-construction communities and completed communities;
[0085] For newly built communities:
[0086] Use a web crawler to search for pre-approval public information online and extract motor vehicle parking space allocation information; the specific steps are as follows:
[0087] (1) Determine the target website: Determine the target website containing information on the construction of motor vehicle parking spaces, which may include official websites of government departments, websites of related industries, or news websites;
[0088] (2) Analyze the web page structure: Use browser developer tools or similar tools to analyze the web page structure of the target website and extract the HTML tags, class names or IDs of the web page;
[0089] (3) Send HTTP request: Use Python's requests library or other web crawler libraries to send HTTP requests to obtain the source code of the target web page;
[0090] (4) Parsing HTML: Use the BeautifulSoup library or other HTML parsing libraries to parse the acquired web page source code; locate the HTML element containing the motor vehicle parking space configuration information based on the previously analyzed web page structure;
[0091] (5) Extracting information: extracting the required information from the parsed HTML, including parking space location, quantity, charging standards, etc.
[0092] (6) Data cleaning and processing: Clean and process the extracted data, remove irrelevant information, and ensure the accuracy and completeness of the data;
[0093] (7) Save data: Save the extracted motor vehicle parking space configuration information to a local file or database for subsequent analysis and use.
[0094] For old and open residential communities, drone aerial video can be used to identify parking spaces using the Canny edge detector in open CV, and then the number of parking spaces can be counted. For communities with missing information, a QR code questionnaire containing information related to the parking census can be filled out in a crowdsourcing manner by the community property management department.
[0095] In order to improve the accuracy of berth census, the intelligent census method of the present invention is also provided with a berth error correction and audit mechanism of static data error correction and dynamic data comparison;
[0096] Among them, static data correction mainly corrects the static berth location, type and other data in the berth census data to check whether there is duplicate or contradictory data.
[0097] Dynamic data comparison is mainly to correct the dynamic berth quantity data in the berth census data. By comparing the existing berth census data and the historical berth census data at the same location, comparing the berth occupancy situation, frequency of use, etc., it is checked whether the berth data changes are too large.
[0098] Establish a regular review mechanism to review the berth census data through regular on-site berth surveys, data comparisons, etc. to ensure the accuracy and consistency of the data.
[0099] Furthermore, the present invention also provides a system for performing data processing and data storage for the above method, including a communication unit for acquiring drone aerial photography and vehicle-mounted video, a database for storing image data, and a processor for processing and identifying images.
[0100] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.
Claims
1. A method for intelligent survey of urban parking resources, characterized in that: This includes a census of on-street public parking spaces, off-street public parking spaces and parking spaces for building accessories; For public parking spaces on the road, the system can intelligently identify the parking fee sign or parking space based on the video and extract the parking information, including the parking location, parking type, parking quantity and parking fee charging method; For off-street public parking spaces, first import the parking space information through the parking lot registration and gate system. If the information cannot be imported, the parking space information is obtained by intelligently identifying the toll sign or parking space based on the video. For the parking spaces built with buildings, including newly built communities and open old communities, the newly built communities mainly use web crawlers to search for pre-approval public information online, extract the parking space construction information for motor vehicles, and determine the parking space information for the newly built communities, including the following steps: (1) Determine the target website containing information on motor vehicle parking spaces and analyze the web page structure of the corresponding website to extract the HTML tags, class names or IDs of the web pages; (2) Use the web crawler library to send HTTP requests to obtain the source code of the target web page and parse the HTML to locate the HTML element containing the motor vehicle parking space accessories information, and then complete the extraction of information, including the parking space location, quantity, and charging standards; (3) Clean and save the data; The number of parking spaces in the open old residential area is identified and processed through drone aerial video and image detection technology, and then the number of parking spaces is counted.
2. The method for intelligent survey of urban parking resources according to claim 1, characterized in that: For on-street public parking spaces, video-based parking location recognition mainly trains the image classification and recognition algorithm by building a parking location map database, then identifies the specific parking location, and further determines the parking type based on the parking location.
3. The method for intelligent survey of urban parking resources according to claim 1, characterized in that: For on-street public parking spaces and off-street public parking spaces, the steps for identifying the parking space charging status and charging method include: Extract berth image data frame by frame and use target detection algorithm to build berth fee sign recognition model to identify berth fee sign; When no fee sign is identified, the corresponding berth is a free berth; When a toll sign is identified and the identification result is "free", the corresponding berth is a free berth; When there is a fee sign and the recognition result is "fee", the OCR text recognition algorithm is used to further extract the berth fee standard.
4. The method for intelligent survey of urban parking resources according to claim 3, characterized in that: When there is a toll sign, the OCR text recognition algorithm can be used directly to extract the number of parking spaces; when there is no toll sign or the number of parking spaces is not marked on the toll sign, the Canny edge detector in open CV is used to perform contour detection on the preprocessed image, identify the parking space contour, and then identify the parking spaces and count the number of parking spaces.
5. The method for intelligent survey of urban parking resources according to claim 1, characterized in that: The number of parking spaces in the old residential area can be determined by using the Canny edge detector in open CV to perform contour detection on the preprocessed image, identify the parking space contours, and then identify the parking spaces and count the number of parking spaces.
6. The method for intelligent survey of urban parking resources according to claim 1, characterized in that: It also includes a regular review mechanism for berth information, which reviews the berth census data through regular on-site berth surveys and data comparisons. The regular review forms include static data error correction and dynamic data comparison.
7. The method for intelligent survey of urban parking resources according to claim 6, characterized in that: The static data error correction includes correcting the berth location, berth type and charging method in the berth census data to check whether there is duplicate or contradictory data; The dynamic data comparison includes correcting the dynamic berth quantity data in the berth census data, by comparing the existing berth census data and the historical berth census data at the same location, comparing the berth occupancy and usage frequency, and checking whether the berth data changes are too large.
8. The method for intelligent survey of urban parking resources according to claim 1, characterized in that: For old communities with missing information, the community property management department can obtain berth information by scanning the QR code and filling out a questionnaire.
9. A system capable of executing the method according to any one of claims 1 to 8, characterized in that: It includes a communication unit for acquiring drone aerial photography and vehicle-mounted videos, a database for storing image data, and a processor for processing and identifying images.
