Urban rainfall distributed monitoring system and method based on traffic participant behaviors

Abstract

The invention provides an urban rainfall distributed monitoring system and method based on traffic participant behaviors, and belongs to the technical field of urban meteorological monitoring and traffic monitoring. The technical problem that existing rainfall monitoring systems mostly depend on weather stations or rain gauges and are difficult to reflect rainfall differences of all regions of a city in detail is solved. According to the technical scheme, the system comprises a video data acquisition and processing module, a traffic participant behavior recognition module, a proportion calculation and noise elimination module, a smooth processing module and a rainfall intensity matching module. The method has the beneficial effects that behaviors of traffic participants are identified by using monitoring videos, the proportion of a rider wearing a raincoat to a pedestrian holding an umbrella is calculated, and a correlation with rainfall intensity data of a meteorological station is constructed, so that distributed monitoring of the urban rainfall level is realized, and the urban rainfall level is monitored. The method provides a novel and effective solution for urban precipitation level distributed monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of urban meteorological monitoring and traffic surveillance, and in particular to a distributed urban precipitation monitoring system and method based on the behavior of traffic participants. Background Art

[0002] In urban environments, precipitation can affect traffic and public safety in different areas to varying degrees. Existing precipitation monitoring systems mostly rely on weather stations or rain gauges, which are difficult to accurately reflect the precipitation differences in different areas of the city. The behavior of traffic participants, such as the behavior of electric bike riders, such as whether they wear raincoats, and the behavior of pedestrians, such as whether they hold umbrellas, can reflect the precipitation conditions in local areas. Therefore, by monitoring and analyzing the behavior of traffic participants and combining it with precipitation intensity information, the precipitation conditions in the city can be effectively monitored. Summary of the invention

[0003] The purpose of the present invention is to overcome the problems in the background technology and provide a distributed urban precipitation monitoring system and method based on the behavior of traffic participants. The system uses monitoring videos to identify the behavior of pedestrians and cyclists, calculates the ratio of the number of electric bike riders wearing raincoats and pedestrians holding umbrellas, and combines precipitation intensity data to monitor precipitation conditions in different areas of the city.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is specifically: a distributed urban precipitation monitoring system based on traffic participant behavior, including the following modules:

[0005] Video data acquisition and processing module, used to collect and process real-time video data of cyclists and pedestrians through cameras;

[0006] Traffic participant behavior recognition module, used to identify the behaviors of traffic participants in the video and classify and count them;

[0007] The ratio calculation and noise removal module is used to calculate the ratio of cyclists wearing raincoats and pedestrians holding umbrellas, and to remove data from periods when the number of electric vehicles and pedestrians is too small;

[0008] The smoothing processing module is used to perform time series smoothing on the proportion data after noise removal to eliminate random fluctuations in the previous and next frame data;

[0009] The precipitation intensity matching module is used to construct the correlation between the smoothed ratio of people wearing raincoats and holding umbrellas and the precipitation intensity of the meteorological station, forming a precipitation intensity matching mechanism.

[0010] The present invention also provides a method for distributed monitoring of urban precipitation, comprising the following steps:

[0011] Step S1: Video data collection and processing: collect real-time video data through urban road monitoring cameras, capture information of cyclists and pedestrians, and create a data set;

[0012] Step S2: Traffic participant behavior recognition: using computer vision algorithms to identify the number of cyclists and pedestrians in the video, and to record and track their numbers. Using image processing technology, it can automatically identify whether cyclists are wearing raincoats and pedestrians are holding umbrellas.

[0013] Step S3: proportion calculation and noise elimination, calculating the proportion of cyclists wearing raincoats and pedestrians holding umbrellas, and eliminating noise data during periods when the number of traffic participants is too small;

[0014] Step S4: Perform time series smoothing on the noise-removed proportional data to eliminate random fluctuations in the previous and next frame data and ensure data stability;

[0015] Step S5: Precipitation intensity matching: matching the smoothed proportional data with the precipitation intensity data of the meteorological station to generate a corresponding relationship between precipitation intensity and traffic participant behavior, and then describing the urban precipitation intensity through proportional data.

[0016] Furthermore, in step S1, the video data acquisition and processing includes the following steps:

[0017] Step 11: Collect video data of the precipitation period on the main traffic arteries and crowded areas of the city through urban road surveillance cameras;

[0018] Step 12: Using a video slicing method, extract the video period with more obvious precipitation, that is, the period when cyclists start to wear raincoats and pedestrians start to hold umbrellas, and slice the video data of this period into image data;

[0019] Step 13: Use annotation tools to annotate the image data, including the location and category of cyclists and pedestrians;

[0020] Step 14: Organize the image data to form a data set, and divide it into training set, validation set and test set.

[0021] Furthermore, in step 12, the video slicing method is implemented by audio and video processing or related tools, the audio and video processing tool reads the input video data and analyzes the video stream and audio stream, further segments the video data of several periods of obvious precipitation, and then decodes and frames the sliced ​​video data, and converts each frame into a required image format;

[0022] The relevant tools mentioned are FFmpeg or Python.

[0023] Furthermore, in step 13, the use of the annotation tool to annotate the image data is achieved by using the annotation tool to select the cyclists and pedestrians in the image, annotate the corresponding labels, select the saving format of the annotation results, organize the annotated images, and divide the data set;

[0024] The labeling tool is labelImg, labelme or Vott.

[0025] Furthermore, in step S2, the traffic participant behavior recognition specifically includes the following steps:

[0026] Step 21: Build a target detection model through YOLO, and use the established data set to train and verify the model;

[0027] Step 22: Adjust the training strategy based on the recall rate and F1-score of the model;

[0028] Step 23: Deploy the trained model to detect real-time video streams or static images. The model outputs the location, size, and classification label of each cyclist and pedestrian, and counts the detection boxes as the number of cyclists and pedestrians.

[0029] Furthermore, in step S3, the ratio calculation and noise removal specifically include the following steps:

[0030] Step 31: According to the number of detection frames of different classification labels output by the model in step 23, calculate the proportion of cyclists wearing raincoats and pedestrians holding umbrellas. The calculation formula is as follows:

[0031]

[0032] in:

[0033] P: is the proportion of cyclists wearing raincoats,

[0034] y: is the number of cyclists wearing raincoats,

[0035] Y: is the number of cyclists,

[0036] m: is the set electric vehicle quantity threshold,

[0037]

[0038] in:

[0039] Q: What is the proportion of pedestrians holding umbrellas?

[0040] z: the number of people holding umbrellas for pedestrians,

[0041] Z: is the number of pedestrians,

[0042] n: is the set pedestrian number threshold,

[0043] Step 32: According to the threshold value of the number of traffic participants set by the system, the proportion data when the number of traffic participants in the time period is less than the set value is eliminated.

[0044] Furthermore, in step S4, the time series smoothing process is performed on the noise-removed proportional data, specifically including the following steps:

[0045] Step 41: Collecting traffic participant behavior ratio data after noise removal;

[0046] Step 42: Calculate the moving average based on the selected window size N, using the following formula:

[0047]

[0048] in:

[0049] S t : is the moving average at time point t,

[0050] x t : is the proportion data at time point t,

[0051] N: is the window size;

[0052] Step 43: Verify the smoothed data by dividing the time series into a training set and a test set, perform smoothing on the training set, and then verify the smoothing effect on the test set.

[0053] Furthermore, in step S5, the precipitation intensity matching includes the following steps:

[0054] Step 51: Collect precipitation data with the same timestamp as the video data, ensure the consistency of the time format, remove outliers and missing values, so as to match with the proportional data;

[0055] Step 52: Match the processed data of the proportion of cyclists wearing raincoats, the proportion of pedestrians holding umbrellas, and the precipitation intensity according to the timestamp, and integrate the matched data into a new data frame including the time, the proportion of cyclists wearing raincoats, the proportion of pedestrians holding umbrellas, and the precipitation intensity;

[0056] Use data software Excel or Python to integrate data and perform model fitting;

[0057] Step 53: Regression analysis is performed on the precipitation data and the proportion of people wearing raincoats, and the precipitation data and the proportion of pedestrians holding umbrellas, respectively, to obtain regression results, i.e., fitting curves and expressions;

[0058] According to the integrated data, a scatter plot of the proportion data and precipitation intensity is drawn, so that the relationship between the two can be observed intuitively and the corresponding regression model is selected. Based on the fact that the proportion data and precipitation data have the same change trend, taking univariate linear regression as an example, its expression is:

[0059] y=ax+b

[0060] in:

[0061] y: precipitation intensity,

[0062] x: is the proportional data,

[0063] a: is the slope, which indicates the expected change in the dependent variable when the independent variable increases by one unit.

[0064] b: intercept, which indicates the expected value of the dependent variable when the independent variable is zero;

[0065] Step 54: Analyze the regression results and calculate R 2 Statistical indicators such as values ​​are used to judge the goodness of fit of the model. The calculation formula is:

[0066]

[0067] in:

[0068] R 2 : is the coefficient of determination, which is used to test the goodness of fit of the model to the data.

[0069] is the predicted value of precipitation intensity,

[0070] y i : is the true value of precipitation intensity,

[0071] is the average value of the true value of precipitation intensity;

[0072] Step 55: Based on the correlation between the obtained ratio data and the precipitation data, the precipitation intensity in different regions is described by the ratio of cyclists wearing raincoats and the ratio of pedestrians holding umbrellas.

[0073] Compared with the prior art, the present invention has the following beneficial effects:

[0074] 1. The present invention uses surveillance video to identify the behavior of traffic participants, calculates the ratio of cyclists wearing raincoats to pedestrians holding umbrellas, and builds a correlation with the precipitation intensity data of the weather station to achieve distributed monitoring of urban precipitation levels. In order to improve data accuracy, the system also introduces a noise removal mechanism to filter out data from time periods with insufficient number of participants, and eliminates random fluctuations through time series smoothing, thereby ensuring the reliability of monitoring results. This method effectively overcomes the problem that most existing precipitation monitoring systems rely on weather stations or rain gauges, making it difficult to carefully reflect the precipitation differences in various areas of the city.

[0075] 2. Accurate precipitation monitoring. By analyzing the behavior of traffic participants, such as cyclists wearing raincoats and pedestrians holding umbrellas, the system can more accurately reflect the precipitation conditions in local areas of the city and capture subtle changes that traditional meteorological monitoring methods may overlook, especially in complex urban environments.

[0076] 3. Real-time data collection and analysis: Using surveillance videos to identify the behavior of traffic participants in real time can quickly collect precipitation-related data. This real-time analysis capability enables city managers to understand precipitation conditions in a timely manner, respond quickly, and improve the city’s ability to respond to sudden weather events.

[0077] 4. Regional monitoring: The system can monitor precipitation in different areas of the city, which helps to identify the uneven distribution of precipitation. This is of great significance for formulating regional flood control and drainage strategies, and can help optimize the planning and resource allocation of urban infrastructure.

[0078] 5. Improved data accuracy. The introduction of noise removal mechanism and time series smoothing can effectively reduce the impact of interference factors on the data and improve the accuracy and reliability of monitoring data, which is crucial for subsequent data analysis and decision support. The system not only improves the accuracy and efficiency of urban precipitation monitoring, but also provides new ideas and methods for urban management, which is of great significance to enhancing the city's ability to cope with climate change. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0080] Figure 1 It is a system structure flow chart of the present invention. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0082] Example

[0083] This embodiment provides a distributed urban precipitation monitoring system based on traffic participant behavior, which specifically includes the following modules:

[0084] Video data acquisition and processing module, used to collect and process real-time video data of cyclists and pedestrians through cameras;

[0085] Traffic participant behavior recognition module, used to identify the behaviors of traffic participants in the video and classify and count them;

[0086] The ratio calculation and noise removal module is used to calculate the ratio of cyclists wearing raincoats and pedestrians holding umbrellas, and to remove data during periods when there are too few traffic participants;

[0087] The smoothing processing module is used to perform time series smoothing on the proportion data after noise removal to eliminate random fluctuations in the previous and next frame data;

[0088] The precipitation intensity matching module is used to construct the correlation between the smoothed ratio of people wearing raincoats and holding umbrellas and the precipitation intensity of the meteorological station, forming a precipitation intensity matching mechanism.

[0089] like Figure 1 As shown, in order to better achieve the above invention effects, the present invention also provides a distributed monitoring method for urban precipitation, and the specific working principle and steps are as follows:

[0090] Step S1: Video data collection and processing: Collect real-time video data through urban road monitoring cameras, capture information of traffic participants, i.e., electric vehicle riders and pedestrians, and produce the required data set;

[0091] S11: Collect video data of the precipitation period on the main traffic arteries and crowded areas of the city through urban road surveillance cameras;

[0092] In S11, the system collects information about traffic participants in real time through cameras installed at major intersections and densely populated areas in the city. The cameras should have night shooting capabilities so that effective video data can be obtained under different lighting conditions. The data is transmitted to the central server through the network for processing. The selection of major intersections and densely populated areas in the city is intended to improve the breadth and reliability of the data set;

[0093] S12: using a video slicing method, extracting a video period with more obvious precipitation, that is, a period when cyclists start to wear raincoats and pedestrians start to hold umbrellas, and slicing the video data of this period into image data;

[0094] In S12, the video slicing may be implemented by audio and video processing or related tools such as FFmpeg or Python. The audio and video processing tool reads the input video data and analyzes the video stream and audio stream, further segments the video data of several periods of obvious precipitation, and then decodes and frames the sliced ​​video data to convert each frame into a required image format.

[0095] S13: Use annotation tools to annotate the image data, including the location and category of traffic participants, i.e., cyclists and pedestrians;

[0096] Image annotation in S13 can be achieved through annotation tools, such as labelImg, labelme, or Vott. Use the annotation tools to select cyclists and pedestrians in the image, and label them with corresponding labels such as "electric vehicle", "wearing raincoat", "pedestrian", "holding umbrella", etc., and select the format of saving the annotation results, YOLO, XML, or PASCAL VOC, organize the annotated images, divide the data set, and convert the data set format if necessary;

[0097] S14: Organize the image data to form a data set, and divide it into a training set, a validation set, and a test set.

[0098] Step S2, traffic participant behavior recognition: using computer vision algorithms to recognize the number of electric vehicles and pedestrians in the video, and recording and tracking the number, and using image processing technology to automatically recognize whether the electric vehicle rider is wearing a raincoat and whether the pedestrian is holding an umbrella;

[0099] S21: Build an object detection model, such as YOLO or Faster R-CNN, and use the established dataset to train and verify the model;

[0100] The target detection model in S21 takes YOLOv8 as an example, and its principle is as follows:

[0101] First, image preprocessing is performed. YOLOv8 uses Letterbox The resize method adjusts the image to the size required by the model, keeps the original image's aspect ratio and scales it proportionally and fills the edges, then normalizes the image so that its pixel values ​​are scaled to between 0 and 1; secondly, the forward propagation of the model, the Backbone of the model extracts features from the processed image, these features include the contour features of cyclists and pedestrians, such as the shape of people, umbrellas, electric vehicles, etc., texture features, such as the texture of clothes, the texture of electric vehicle frames, etc., color features, such as the color of raincoats, the color of electric vehicles, etc.; then Neck fuses feature maps of different scales, for example, extracts feature maps of different scales at different layers of the network, and then fuses these feature maps, so that the model can detect both smaller cyclists in the distance and larger cyclists in the near distance, thereby enhancing the expressive power of the features; then the output is passed to the Head part for the final prediction, and the prediction results include the bounding boxes and categories of traffic participants; finally, post-processing, the output prediction results are threshold filtered, that is, the results with confidence levels lower than the preset threshold are eliminated, and the NMS, i.e. non-maximum suppression, operation is used to remove overlapping detection boxes and retain the best detection results;

[0102] S22: Adjust the training strategy in time according to the model's recall rate, F1-score and other indicators;

[0103] S23: Deploy the trained model to detect real-time video streams or static images. The model outputs the location, size, and classification labels of each cyclist and pedestrian, such as whether they are holding an umbrella or wearing a raincoat, and counts the detection boxes as the number of cyclists and pedestrians.

[0104] Step S3, proportion calculation and noise elimination: calculate the proportion of electric bike riders and people wearing raincoats, and eliminate noise data during periods when the number of traffic participants is too small. The proportion calculation and noise elimination module sets a threshold for the number of electric bikes. When the number of electric bikes in a certain time period is lower than the set threshold, for example, 5, the data for this period will be eliminated to prevent noise from affecting the accuracy of data statistics.

[0105] S31: According to the number of detection frames of different classification labels output by the model in step S2, the proportion of cyclists wearing raincoats and the proportion of pedestrians holding umbrellas are calculated. The calculation formula is as follows:

[0106]

[0107] in:

[0108] P: is the proportion of cyclists wearing raincoats,

[0109] y: is the number of cyclists wearing raincoats,

[0110] Y: is the number of cyclists,

[0111] m: is the set electric vehicle quantity threshold,

[0112]

[0113] in:

[0114] Q: What is the proportion of pedestrians holding umbrellas?

[0115] z: the number of people holding umbrellas for pedestrians,

[0116] Z: is the number of pedestrians,

[0117] n: is the set pedestrian number threshold;

[0118] S32: According to the threshold value of the number of traffic participants set by the system, when Y≤m or Z≤n, the proportion data when the number of traffic participants in the time period is less than the set value is eliminated.

[0119] Step S4, in the smoothing processing module: performing time series smoothing processing on the noise-removed ratio data to eliminate random fluctuations in the previous and next frame data to ensure data stability, that is, using the moving average method or the weighted smoothing algorithm in the smoothing processing module to smooth the time series data of the ratio of the number of people wearing raincoats to eliminate random fluctuations in a short period of time and ensure data stability;

[0120] S41: Collect traffic participant behavior ratio data after noise removal;

[0121] S42: Calculate the moving average according to the selected window size N, that is, the time period or number of image frames considered when calculating the average value, and the formula is as follows:

[0122]

[0123] in:

[0124] S t : is the moving average at time point t,

[0125] x t : is the proportion data at time point t,

[0126] N: is the window size;

[0127] S43: Verify the smoothed data, that is, divide the time series into a training set and a test set, perform smoothing on the training set, and then verify the smoothing effect on the test set.

[0128] Step S5, matching precipitation intensity: matching the smoothed ratio of people wearing raincoats and the ratio of pedestrians holding umbrellas with the precipitation intensity data of the meteorological station, and then describing the precipitation intensity of the city through the ratio data, that is, matching the smoothed ratio data of people wearing raincoats with the precipitation intensity data provided by the meteorological station through the precipitation intensity matching module, so as to generate a corresponding relationship between precipitation intensity and the behavior of traffic participants;

[0129] S51: Collect precipitation data with the same timestamp as the video data, ensure the consistency of the time format, and remove outliers and missing values ​​to match with the proportional data;

[0130] S52: matching the processed data of the proportion of people wearing raincoats, the proportion of pedestrians holding umbrellas, and the precipitation intensity according to the timestamp, and integrating the matched data into a new data frame including the time, the proportion of people wearing raincoats, the proportion of pedestrians holding umbrellas, and the precipitation intensity;

[0131] You can use statistical software such as Excel or Python to integrate the data and perform model fitting;

[0132] S53: performing regression analysis on the precipitation data and the proportion of people wearing raincoats, and the precipitation data and the proportion of pedestrians holding umbrellas, respectively, to obtain regression results, that is, to obtain fitting curves and expressions;

[0133] According to the integrated data, a scatter plot of the proportion data and precipitation intensity is drawn, so that the relationship between the two can be observed intuitively and the corresponding regression model is selected. Based on the fact that the proportion data and precipitation data have the same change trend, taking univariate linear regression as an example, its expression is:

[0134] y=ax+b

[0135] in:

[0136] y: precipitation intensity,

[0137] x: is the proportional data,

[0138] a: is the slope, which indicates the expected change in the dependent variable when the independent variable increases by one unit.

[0139] b: intercept, which indicates the expected value of the dependent variable when the independent variable is zero;

[0140] S54: Analyze regression results and calculate R 2 Statistical indicators such as values ​​are used to judge the goodness of fit of the model. The calculation formula is:

[0141]

[0142] in:

[0143] R 2: is the coefficient of determination, which is used to test the goodness of fit of the model to the data.

[0144] is the predicted value of precipitation intensity,

[0145] y i : is the true value of precipitation intensity,

[0146] is the average value of the true value of precipitation intensity,

[0147] R 2 Used to evaluate the goodness of fit of the model, R 2 The closer the value is to 1, the better the model fit;

[0148] S55: According to the correlation between the obtained ratio data and the precipitation data, the precipitation intensity in different regions can be described by the ratio of cyclists wearing raincoats and the ratio of pedestrians holding umbrellas.

[0149] In summary, by analyzing the behaviors of electric bike riders wearing raincoats and pedestrians holding umbrellas, more accurate precipitation monitoring can be achieved in different areas within the city. This behavior-based monitoring method overcomes the limitations of traditional weather stations and rain gauges, and can capture the precipitation differences within the city in detail. At the same time, surveillance video technology is used to identify the behavior of traffic participants in real time, and combined with precipitation intensity data, dynamic distributed monitoring of precipitation conditions can be achieved. This real-time nature provides timely information support for urban management and emergency response. In addition, the system introduces a noise removal mechanism, which can filter out data for time periods with insufficient number of participants, and eliminate random fluctuations through time series smoothing. This measure significantly improves the reliability of monitoring results and makes data analysis more accurate.

[0150] The present invention provides an urban precipitation monitoring system and method based on the behavior of traffic participants. By real-time monitoring of the behavior of traffic participants, such as cyclists wearing raincoats and pedestrians holding umbrellas, and combining with meteorological station data, the precipitation conditions in local areas of the city can be effectively monitored. The system has high accuracy, real-time and regionality, can provide effective precipitation data support for urban management, and help optimize infrastructure planning and resource allocation.

[0151] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed urban precipitation monitoring system based on traffic participant behavior, characterized in that: Includes the following modules: Video data acquisition and processing module, used to collect and process real-time video data of cyclists and pedestrians through cameras; Traffic participant behavior recognition module, used to identify the behaviors of traffic participants in the video and perform classification and statistics; The ratio calculation and noise removal module is used to calculate the ratio of cyclists wearing raincoats and pedestrians holding umbrellas, and to remove data from periods when the number of electric vehicles and pedestrians is too small; The smoothing processing module is used to perform time series smoothing on the proportion data after noise removal to eliminate random fluctuations in the previous and next frame data; The precipitation intensity matching module is used to construct the correlation between the smoothed ratio of people wearing raincoats and holding umbrellas and the precipitation intensity of the meteorological station, forming a precipitation intensity matching mechanism.

2. A distributed monitoring method for urban precipitation, the method is based on the distributed monitoring system for urban precipitation based on traffic participant behavior as claimed in claim 1, characterized in that: The steps include: Step S1: Video data collection and processing: collect real-time video data through urban road monitoring cameras, capture information of cyclists and pedestrians, and create a data set; Step S2: Traffic participant behavior recognition: using computer vision algorithms to identify the number of cyclists and pedestrians in the video, and to record and track their numbers. Using image processing technology, it can automatically identify whether cyclists are wearing raincoats and pedestrians are holding umbrellas. Step S3: proportion calculation and noise elimination, calculating the proportion of cyclists wearing raincoats and pedestrians holding umbrellas, and eliminating noise data during periods when the number of traffic participants is too small; Step S4: Perform time series smoothing on the noise-removed proportional data to eliminate random fluctuations in the previous and next frame data and ensure data stability; Step S5: Precipitation intensity matching: matching the smoothed proportional data with the precipitation intensity data of the meteorological station to generate a corresponding relationship between precipitation intensity and traffic participant behavior, and then describing the urban precipitation intensity through proportional data.

3. A distributed monitoring method for urban precipitation according to claim 2, characterized in that: In step S1, the video data collection and processing includes the following steps: Step 11: Collect video data of the precipitation period on the main traffic arteries and crowded areas of the city through urban road surveillance cameras; Step 12: Using a video slicing method, extract the video period with more obvious precipitation, that is, the period when cyclists start to wear raincoats and pedestrians start to hold umbrellas, and slice the video data of this period into image data; Step 13: Use annotation tools to annotate the image data, including the location and category of cyclists and pedestrians; Step 14: Organize the image data to form a data set, and divide it into training set, validation set and test set.

4. The method for distributed monitoring of urban precipitation according to claim 3, characterized in that: In step 12, the video slicing method is implemented by audio and video processing or related tools. The audio and video processing tool reads the input video data and analyzes the video stream and audio stream, further segments the video data of several periods of obvious precipitation, and then decodes and frames the sliced ​​video data to convert each frame into the required image format; The relevant tools mentioned are FFmpeg or Python.

5. The method for distributed monitoring of urban precipitation according to claim 3, characterized in that: In step 13, the image data is annotated using an annotation tool, which is implemented by using the annotation tool to select cyclists and pedestrians in the image, annotate corresponding labels, select a saving format for the annotation results, organize the annotated images, and divide the data set; The labeling tool is labelImg, labelme or Vott.

6. A distributed monitoring method for urban precipitation according to claim 5, characterized in that: In step S2, the traffic participant behavior recognition specifically includes the following steps: Step 21: Build a target detection model through YOLO, and use the established data set to train and verify the model; Step 22: Adjust the training strategy based on the recall rate and F1-score of the model; Step 23: Deploy the trained model to detect real-time video streams or static images. The model outputs the location, size, and classification label of each cyclist and pedestrian, and counts the detection boxes as the number of cyclists and pedestrians.

7. A distributed monitoring method for urban precipitation according to claim 6, characterized in that: In step S3, the ratio calculation and noise removal specifically include the following steps: Step 31: According to the number of detection frames of different classification labels output by the model in step 23, calculate the proportion of cyclists wearing raincoats and pedestrians holding umbrellas. The calculation formula is as follows: in: P: is the proportion of cyclists wearing raincoats, y: is the number of cyclists wearing raincoats, Y: is the number of cyclists, m: is the set electric vehicle quantity threshold, in: Q: What is the proportion of pedestrians holding umbrellas? z: the number of people holding umbrellas for pedestrians, Z: is the number of pedestrians, n: is the set pedestrian number threshold, Step 32: According to the threshold value of the number of traffic participants set by the system, the proportion data when the number of traffic participants in the time period is less than the set value is eliminated.

8. The method for distributed monitoring of urban precipitation according to claim 7, characterized in that: In step S4, the time series smoothing process is performed on the noise-removed proportional data, which specifically includes the following steps: Step 41: Collecting traffic participant behavior ratio data after noise removal; Step 42: Calculate the moving average based on the selected window size N, using the following formula: in: S t : is the moving average at time point t, x t : is the proportion data at time point t, N: is the window size; Step 43: Verify the smoothed data by dividing the time series into a training set and a test set, perform smoothing on the training set, and then verify the smoothing effect on the test set.

9. A distributed monitoring method for urban precipitation according to claim 8, characterized in that: In step S5, the precipitation intensity matching includes the following steps: Step 51: Collect precipitation data with the same timestamp as the video data, ensure the consistency of the time format, remove outliers and missing values, so as to match with the proportional data; Step 52: Match the processed data of the proportion of cyclists wearing raincoats, the proportion of pedestrians holding umbrellas, and the precipitation intensity according to the timestamp, and integrate the matched data into a new data frame including the time, the proportion of cyclists wearing raincoats, the proportion of pedestrians holding umbrellas, and the precipitation intensity; Use data software Excel or Python to integrate data and perform model fitting; Step 53: Regression analysis is performed on the precipitation data and the proportion of people wearing raincoats, and the precipitation data and the proportion of pedestrians holding umbrellas, respectively, to obtain regression results, i.e., fitting curves and expressions; Step 54: Analyze the regression results and calculate R 2 The statistical index of the value is used to judge the goodness of fit and significance of the model. The calculation formula is: in: R 2 : is the coefficient of determination, which is used to test the goodness of fit of the model to the data. is the predicted value of precipitation intensity, y i : is the true value of precipitation intensity, is the average value of the true value of precipitation intensity; Step 55: Based on the correlation between the obtained ratio data and the precipitation data, the precipitation intensity in different regions is described by the ratio of cyclists wearing raincoats and the ratio of pedestrians holding umbrellas.

Images