Street safety perception evaluation and optimization method based on multi-source data
By integrating multi-source data and deep learning technology, a security perception assessment module is built, which solves the problem of difficulty in quantifying street security perception in the existing technology, and accurately evaluates and optimizes street security.
Patent Information
- Application Number
- CN202510305588.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-14
- Publication Date
- 2025-07-04
AI Technical Summary
The existing technology is difficult to accurately quantify the impact of street environment on individual sense of security, resulting in the inability to provide a scientific basis for street safety planning.
By integrating multi-source data, including road network data, street view image data and three-dimensional building map data, deep learning technology and machine learning algorithms, a security perception assessment module is built to perform security perception assessment and optimization.
It has achieved accurate assessment of the street security perception level and provided scientific optimization strategies to improve street safety.
Smart Images
Figure CN120259052A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of urban planning and environmental design, and particularly to a method for street safety perception evaluation and optimization based on multi-source data. Background Art
[0002] In urban planning and construction, street safety is a crucial consideration factor. With the continuous increase in the public's demand for safety, the traditional street design and management methods are no longer able to meet the requirements. Currently, street safety research faces many challenges, such as the lack of comprehensive and effective data acquisition and analysis methods, making it difficult to accurately quantify the impact of the street environment on individual sense of security, and thus unable to provide a scientific basis for street safety planning. Summary of the Invention
[0003] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide a method for street safety perception evaluation and optimization based on multi-source data. By integrating a variety of data resources and technical means, it accurately evaluates the street safety perception level and provides optimization strategies to improve street safety. To achieve the above object and other advantages according to the present invention, there is provided a method for street safety perception evaluation and optimization based on multi-source data, including:
[0004] S1. Obtain road network data and preprocess the data, where the preprocessing includes screening valid sections and extracting objective street indicators;
[0005] S2. Construct a safety perception evaluation module to perform classification prediction on images. By interval mapping of the prediction confidence, restore the continuous value of the safety perception score to achieve the evaluation of street safety perception;
[0006] S3. Conduct multiple logistic regression analysis on the data, taking the objective street indicators as independent variables and the safety perception scores as dependent variables, analyze the influence degree of each indicator on safety perception, and formulate a scientific and reasonable street safety optimization strategy according to the regression analysis results.
[0007] Preferably, the safety perception evaluation module uses the MIT Spatial Impulse Dataset to guide a large number of users to perform binary marking on street view images, and then is constructed based on the classic deep convolutional neural network ResNet. The safety perception evaluation module performs classification prediction on images. By interval mapping of the prediction confidence, restore the continuous value of the safety perception score to achieve the evaluation of street safety perception.
[0008] A street safety perception evaluation and optimization device based on multi-source data includes: a data collection module for collecting map point-of-interest data, street view image data, and three-dimensional building map data of the street; a data processing module for preprocessing the collected data, including screening valid road sections and extracting objective street indicators; a safety perception evaluation module for constructing a safety perception evaluation model using deep learning techniques and machine learning algorithms, combined with the MIT Spatial Pulse dataset, to score the street safety perception; an analysis and optimization module for analyzing the relationship between objective street indicators and safety perception through mathematical derivation, determining the key factors affecting street safety perception, and proposing optimization strategies. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] Figure 1 FIG. is a schematic diagram of the overall structure of the multi-source data-based street safety perception evaluation and optimization method according to the present invention;
[0010] Figure 2 FIG. is a schematic diagram of the method for obtaining the street width of the multi-source data-based street safety perception evaluation and optimization method according to the present invention;
[0011] Figure 3 FIG. is a diagram of using a deep learning model to identify the street greening map of the multi-source data-based street safety perception evaluation and optimization method according to the present invention;
[0012] Figure 4 FIG. is a diagram of using a deep learning model to identify the facade shop signs map of the multi-source data-based street safety perception evaluation and optimization method according to the present invention;
[0013] Figure 5 FIG. is a diagram of the method for predicting human perception of street images of the multi-source data-based street safety perception evaluation and optimization method according to the present invention;
[0014] Figure 6 FIG. is a flowchart of the transparency calculation of the street of the multi-source data-based street safety perception evaluation and optimization method according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0016] Refer to Figures 1-5, A method for street safety perception assessment and optimization based on multi-source data, including: S1. Obtain road network data and preprocess the data. The preprocessing includes screening valid road segments and extracting objective street indicators. The objective street indicators include street width, green view rate, building interface indicators such as the number of store signs, interface permeability, and street function mixing degree. The road network data includes map point of interest (POI) data, street view image data, and 3D building map data of the street.
[0017] Among them, the road segment screening is as follows:
[0018] (1) Obtain road network data from OpenStreetMap. According to the definition of the street and research requirements, screen out the streets that can carry daily social life. Specifically, when operating, delete the community road segments that do not meet the requirements under the viaduct, bridge, in the residential area, etc., to ensure that the research object is the main urban streets.
[0019] (2) Interrupt the screened streets at intersections and remove the 0.5m section at the intersections to avoid the influence of the complex intersection environment on the data accuracy. At the same time, delete the road segments without buildings within a 40m buffer distance to the left and right of the street center line to ensure that the research road segments have a representative street environment.
[0020] (3) Further screen the length of the street segments, retain the road segments with a length of 50m or more, and exclude the interference of too short road segments on the overall analysis. In addition, according to past calculation experience, delete the road segments with too few street view sampling points (less than 4) and those that cannot obtain reasonable street view-related data, and finally determine the valid road segments.
[0021] The acquisition of street view images is as follows:
[0022] (1) Obtain street view image data from Baidu Street View Service through the API. After manual comparison and analysis, determine the sampling distance to be 25m. This distance can ensure the diversity of landscape types covered by the images while avoiding excessive overlap or omission of landscape features.
[0023] (2) Request street view images according to the sampling point positions, set the image size to 1024×512 pixels, the camera compass heading to 0°, 90°, 180°, 270°, and the horizontal field of view of the image to 90°. Such parameter settings can capture the environmental information around the sampling points in all directions.
[0024] 3. Acquisition of map point of interest (POI) data
[0025] Collect the summary information of 9 types of POIs in Amap data, including functional types such as residence, commerce, transportation, catering and entertainment, education and training, medical services, enterprise factory offices, government agencies and social organizations, and green spaces. These data will be used for the subsequent calculation of street function density and function mixing degree.
[0026] The data is processed as follows:
[0027] 1. Preprocessing of street view images
[0028] (1) Clean the collected street view images to remove noise, blurred parts, and interference elements unrelated to street safety perception, such as flying birds in the sky, distant unrelated buildings, etc., to improve the quality of image data.
[0029] (2) Standardize the images by unifying parameters such as brightness, contrast, and color saturation of the images, making the images collected under different times and weather conditions comparable, facilitating subsequent analysis and processing.
[0030] 2. Processing of 3D building map data
[0031] (1) Geometric correction of the 3D building map data to correct problems such as building position deviation and shape deformation that may occur during data collection, ensuring the accuracy of building position and shape.
[0032] (2) Perform spatial registration to precisely match the 3D building map data with street view image data, road network data, etc., so that the data from different data sources is consistent in space, providing a basis for comprehensive analysis.
[0033] The objective indicators of the street are obtained as follows:
[0034] 1. Calculation of street width
[0035] (1) Use the centerline and building base vector information collected by GIS to create a series of buffers at intervals of 1m steps outward from the centerline (the maximum width is 40m).
[0036] (2) Calculate the area ratio of the unilateral buffer area and the area after removing the building base, and determine the side interface line by querying the mutation point with the largest difference in the ratio between adjacent two buffers.
[0037] (3) Finally, calculate the sum of the distances from the interface lines on both sides of the road section to the centerline to obtain the street width.
[0038] The method for obtaining the street width is as Figure 2 shown:
[0039] The calculation content and description of the transparency of the street are as shown in the following table and Figure 6 shown:
[0040]
[0041] 2. Measurement of green view rate
[0042] (1) Apply the deep learning model PSPNet. With the support of DCNN, this model can assign a class label to each pixel in the image, and the pixelization accuracy reaches 79.70%.
[0043] (2) Use this model to identify various vegetation in street view images, including trees, shrubs, flowers, vegetables, etc., and count the pixel proportion of the vegetation, which is used as the green view rate of this sampling point. The average green view rate of the sampling points on each street section is the green view rate of this road section.
[0044] 3. Statistical count of the number of storefront signs
[0045] (1) Use the object detection machine learning algorithm to identify the storefront signs in street view images. This algorithm can achieve high accuracy through training with a small number of samples and is suitable for identifying storefront signs in complex street environments.
[0046] (2) Calculate the average value of the number of storefront signs at all sampling points in the street section, which is recorded as the average number of storefront signs per sampling point on this road section, with the unit of number / sampling point. This can avoid the quantity differences caused by different sampling lengths and total numbers of sampling points on each road section and accurately reflect the average number of visible storefront signs at a single sampling point on this road section.
[0047] 4. Calculation of the interface permeability
[0048] (1) Manually calibrate to obtain the training set data, establish a deep learning image recognition model, and use semantic scene parsing technology to interpret the elements of the Shanghai street space.
[0049] (2) Input the image data of all points, and identify the permeable interface elements, such as the openings of the streets, glass surfaces, display windows, overhead corridors, perforated fences, open entrances to public areas, etc. (see the relevant table for specific contents).
[0050] (3) Summarize the element compositions corresponding to the east, west, south, and north 4 directions of each street point respectively, and calculate the proportion of the transparent interface in the building interface in each street view image at this point, with the value expressed as a percentage.
[0051] 5. Calculation of the street function density
[0052] (1) Select the POI points related to vitality within 55m on both sides of the street (which can include the street map interest point positions) for statistical analysis.
[0053] (2) Calculate the function density of the street according to the formula "function density = total number of functions / street length × 100", with the unit of number / km. The specific function types are divided into 9 types: residential, commercial, transportation, catering and entertainment, education and training, medical services, enterprise factory offices, government agencies and social organizations, and green spaces. By counting the number of POI points of these function types, the function density of the street can be obtained.
[0054] 6. Calculation of street function mixing degree
[0055] (1) Assume that the total number of POIs around a street is A, which is divided into N types in total (N ≤ 9).
[0056] If the numbers of each type are A1, A2, ……, A n ;
[0057] (2) Then the calculation formula is:
[0058] (3) Define the probability as P i , and the calculation formula is:
[0059] (4) Obviously, there is Thus, the information entropy H of the street function is obtained, and the calculation formula is:;
[0060] (5)
[0061] S2. Build a safety perception evaluation module to classify and predict images. Through the interval mapping of the prediction confidence level, restore the continuous value of the safety perception score to realize the evaluation of street safety perception; specifically as follows:
[0062] 1. Data marking and model training
[0063] (1) Use the spatial impulse collection platform of the Massachusetts Institute of Technology to let a large number of users perform binary marking on a dataset containing 110,988 street view images taken between 2007 and 2012 (in 56 cities across 28 countries on 6 continents). Users choose the image they think is more secure (left or right, or "equally secure") from two randomly selected street view pictures to indicate their perceptual judgment.
[0064] (2) Participants were asked to choose one of the two randomly selected street view pictures that they thought was more secure. Each image sample i was compared with other image i′ multiple times. Define the positive rate (P) of image i along a certain perception index as: The negative rate (N) is:
[0065] In the formula: p i and n i respectively represent the number of times image i is selected and not selected in the comparison, and e i represents the number of times image i is considered equal to another image in the comparison. The calculation formula for the score Q of picture i is:
[0066] (3) Based on the classic deep convolutional neural network ResNet, a model is built between street view images and individual perception. The street view images are input into the model, and the deep convolutional neural network DCNN and the SVM classifier are used to perform binary classification prediction on the images. Then, interval mapping is performed on the predicted probability scores to restore the continuous values of the safety perception scores, realizing the scoring of street safety perception. The overall experiment achieves an accuracy of 72% in the prediction of the sense of security.
[0067] 2. Calculation of safety perception score
[0068] Each picture is automatically scored through the above model, and the average score of each picture is statistically calculated by section. The total score of the section is on a 10-point scale, from 0 to 10 points. The higher the score, the higher the sense of security.
[0069] S3. Conduct multiple logistic regression analysis on the data, taking the objective indicators of the street as independent variables and the safety perception score as the dependent variable, analyzing the influence degree of each indicator on safety perception. According to the regression analysis results, formulate scientific and reasonable street safety optimization strategies.
[0070] 1. Among them, multiple logistic regression analysis
[0071] (1) Regard 6 indicators of 3 categories of the street (street width, green view rate, number of store signs, interface permeability, function density, function mixing degree) as independent variables and the safety perception score as the dependent variable, and use linear regression analysis to analyze the influence of the street indicators at the sampling points on street safety perception. The formula is as follows: SAFETY = a0 + a1×SHOPSIGN + a2×W + a3×GREEN + a4×PERMEABILITY + a5×POI + a6×∑ NI + ε; where: a1 - a6 are weight coefficients, a0 is a constant term, ε is a random error, SAFETY represents the subjective score of safety perception, W represents the street width, GREEN represents the green view rate, PERMEABILITY represents the permeability, SHOPSIGN represents the number of store signs, POI represents the function density, and ∑ NI represents the function mixing degree.
[0072] (2) Analyze the regression results. All elements pass the significance test at the 0.05 level, and the VIF value after regression is less than 5, indicating that there is no collinearity between the independent variables. The regression model passes the hypothesis test, and R2 is 0.524, with a good fitting degree.
[0073] 2. Estimation of appropriate values
[0074] According to the regression analysis results and scatter plot analysis, determine the appropriate value ranges of each key factor:
[0075] (1) Street function density: The function density corresponding to the highest safety perception value of 10 points is about 1500 per km. Controlling the appropriate street function density within the range of 300 - 1200 is beneficial to street safety perception.
[0076] (2) Street function mixing degree: When the function mixing degree is between 1.2 - 2.2, the safety perception scores are generally high. When it is less than 1, the impact on the sense of security is not significant. It is necessary to ensure that the function mixing degree reaches a certain numerical range.
[0077] (3) Average number of store signs: Controlling the appropriate street function density within the range of 10 - 30 per sampling point is beneficial to street safety perception.
[0078] (4) Permeability: The permeability of the interface with a perception score above 6 - 8 points is between 5% - 15%. When the permeability is greater than 60%, the safety level gradually decreases as the green view rate increases. The appropriate value range is 20% - 55%.
[0079] (5) Green view rate: The highest safety level appears at about 6 points, and the green view rate is 25% - 45% at this time. When the green view rate is greater than 45%, the increase in the sense of security is not obvious. The green view rate for effectively relieving stress is 24% - 34%.
[0080] (6) Street width: The street width from 10 - 80m has little impact on the perception activity. However, for overly wide streets, the safety perception can be improved by dealing with the near - human space part and arranging diverse activities.
[0081] A street safety perception evaluation and optimization device based on multi - source data, including:
[0082] A data acquisition module, used to collect map point - of - interest data, street view image data, and three - dimensional building map data of the street;
[0083] A data processing module, used to pre - process the collected data, including screening valid road segments and extracting objective street indicators;
[0084] A safety perception evaluation module, used to score the street safety perception;
[0085] An analysis and optimization module, used to analyze the influence degree of each indicator on the safety perception, and formulate a scientific and reasonable street safety optimization strategy according to the regression analysis results.
[0086] The number of devices and the processing scale described here are used to simplify the description of the present invention. The application, modification, and variation of the present invention are obvious to those skilled in the art.
[0087] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the present invention. For those skilled in the art, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and the equivalent scope, the present invention is not limited to the specific details and the illustrated examples described herein.
Claims
1. A street safety perception evaluation and optimization method based on multi-source data, characterized in that Including: S1. Obtain road network data and preprocess the data, where the preprocessing includes screening valid road segments and extracting objective street indicators; S2. Construct a safety perception evaluation module to classify and predict images. By performing interval mapping on the prediction confidence, restore the continuous value of the safety perception score to achieve the evaluation of street safety perception; S3. Conduct a multiple logistic regression analysis on the data, taking the objective street indicators as independent variables and the safety perception score as the dependent variable, analyze the influence degree of each indicator on safety perception, and formulate a scientific and reasonable street safety optimization strategy according to the regression analysis results.
2. The method for street safety perception evaluation and optimization based on multi-source data according to claim 1, characterized in that The road network data described in step S1 includes map point-of-interest data, street view image data, and three-dimensional building map data of the street.
3. The method for street safety perception evaluation and optimization based on multi-source data according to claim 1, characterized in that, The objective street indicators in step S1 include street width, green view rate, number of store signs in the building interface index, interface permeability, and street function mix degree.
4. The street safety perception evaluation and optimization method based on multi-source data according to claim 3, characterized in that By using the centerline and building base vector information collected by GIS and adopting the buffer analysis method, obtain the street width; Apply the deep learning model PSPNet to directly learn the visual features of green vegetation, accurately identify and count the proportion of vegetation pixels; Adopt an efficient object detection algorithm to calculate the average value of the number of store signs at all sampling points in the road segment; Rely on the deep learning image recognition model established by manually calibrating the training set data to interpret the street spatial elements, accurately identify the permeable interface elements and calculate their proportion in the building interface; By counting the number of POI points related to vitality within a specific range on both sides of the street, calculate the number of functional business types in each 100m street segment according to a specific formula; at the same time, determine the function mix degree by calculating the information entropy and reasonably process the POI categories with less than 5 category totals.
5. The method for street safety perception evaluation and optimization based on multi-source data according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21. Require participants to select one of the two randomly selected street view pictures that the participant believes is more secure; S22. Compare each image sample with other images multiple times, and define the positive rate and negative rate of image i along a certain perception indicator; S23. Calculate the score of the picture.
6. The method for street safety perception evaluation and optimization based on multi-source data according to claim 1, characterized in that In step S3, 3 categories and 6 indicators of the street are regarded as self-indicators, and the safety perception score is regarded as the dependent indicator. Use linear regression analysis to analyze the influence of the street indicators at the sampling points on street safety perception.
7. A street safety perception evaluation and optimization device based on multi-source data, characterized in that, Including: A data acquisition module for collecting map point-of-interest data, street view image data, and three-dimensional building map data of the street; A data processing module for preprocessing the collected data, including screening valid road segments and extracting objective street indicators; A safety perception evaluation module for scoring the street safety perception; An analysis and optimization module for analyzing the influence degree of each indicator on safety perception, and formulating a scientific and reasonable street safety optimization strategy according to the regression analysis results.
8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method described in any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method described in any one of claims 1 to 6.