Safety early warning signal sending control method applied to intelligent monitoring device

Abstract

The application discloses a safety early warning signal sending control method applied to an intelligent monitoring device and belongs to the technical field of artificial intelligence. When a traffic accident occurs, monitoring data, meteorological data and geographical data of the traffic accident occurrence site are acquired to form traffic accident data; visual features, weather features and road environment features are extracted; whether a vehicle violates rules stipulated by traffic signs is identified, and whether the driving track of the vehicle conforms to lane rules is judged to form a traffic violation risk index; in combination with the weather features and the road environment features, an environmental interference intensity index is constituted; damage conditions of traffic signs around the accident site and shielding conditions of buildings on pedestrians are checked to obtain a road facility correlation index; a multi-layer perception model is trained, index data are input, and actual proportions of human factors and natural factors of the traffic accident are output; when the proportion of the natural factors exceeds a threshold value, a reconstruction suggestion of the traffic accident occurrence site is provided.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method for controlling the transmission of safety warning signals in intelligent monitoring equipment. Background Technology

[0002] With the acceleration of urbanization and the continuous increase in the number of motor vehicles, frequent traffic accidents have become a serious social problem, posing a huge impact on people's lives and property safety and social and economic development. Accurate and rapid analysis of the causes of traffic accidents is crucial for efficient accident handling and prevention of similar accidents from happening again.

[0003] Previous methods lacked precision in extracting target features from surveillance videos, only able to identify the general outlines of vehicles and pedestrians, failing to accurately capture key features such as instantaneous vehicle turning angles, changes in pedestrian speed, and detailed status of traffic signs. Existing risk assessment models often focused on single factors, either emphasizing abnormal vehicle trajectories or solely considering the severity of weather conditions, without comprehensively considering the synergistic effects of traffic violations, environmental interference, and road infrastructure conditions. Current technologies primarily focused on post-accident handling, lacking proactive accident prevention capabilities. They could not predict accidents in advance and implement corresponding preventative measures through in-depth analysis of historical accident data and monitoring of potential risk factors. Summary of the Invention

[0004] The purpose of this invention is to provide a method for controlling the transmission of security early warning signals in intelligent monitoring equipment, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for controlling the transmission of security warning signals in intelligent monitoring equipment, the method comprising the following steps:

[0007] When a traffic accident occurs, monitoring data, meteorological data, and geographic data of the accident location are obtained, and the data is preprocessed and integrated to form traffic accident data.

[0008] Based on the traffic accident data, visual features of targets, including vehicles, pedestrians, and traffic signs, are extracted from the monitoring data using the YOLOv5 model; weather features are extracted from meteorological data, and precipitation, wind speed, temperature, and visibility are quantified into feature values ​​that affect the degree of traffic impact; and road environment features are extracted from geographical data.

[0009] Based on the traffic signs, whether the vehicle violates the rules stipulated by the traffic signs is identified, whether the driving trajectory of the vehicle conforms to the lane rules is judged, the operation score of the driver is obtained combined with the weather characteristics, and a traffic violation risk index is formed; combined with the weather characteristics and the road environment characteristics, an environmental interference intensity index is formed; the damage of the traffic signs around the accident site is checked, the positions of the pedestrians and the distribution of the surrounding buildings are obtained, the shielding of the buildings to the pedestrians is checked, and a road facility correlation index is obtained;

[0010] Past traffic accident data is collected, a multi-layer perceptron model architecture is built, the multi-layer perceptron model is trained using the training set data, the weights and biases of the model are continuously adjusted through the back propagation algorithm, the traffic violation risk index, the environmental interference intensity index and the road facility correlation index are taken as inputs, and the actual proportion of human factors and natural factors of the traffic accident is taken as the target output;

[0011] A threshold of the proportion of natural factors is set, when the proportion of natural factors exceeds the threshold, a reconstruction suggestion of the traffic accident location is provided based on the weather characteristics and the road environment characteristics.

[0012] The YOLOv5 model is pre-trained on traffic scene image data, when a traffic accident occurs, the monitoring data in the traffic accident data is input into the YOLOv5 model frame by frame in the image sequence of one frame per second; for vehicles, the vehicle type, body color and license plate number are identified, and the driving trajectory of the vehicle before the accident is tracked; for pedestrians, the relative relationship between the position of the pedestrian at the time of the accident and the vehicle is judged; for traffic signs, the sign type and its position in the image are identified.

[0013] Further, based on the weather characteristics, the precipitation amount is divided into different magnitudes according to meteorological standards, combined with the road environment characteristics, the water depth is estimated, and the working frequency of the wiper, the density of raindrops in front of the lens and the head movement of the driver are observed through the monitoring data to judge the degree of visual obstruction; based on the precipitation level, water depth and visual obstruction degree, the precipitation characteristics are quantified;

[0014] According to the wind speed value in the meteorological data, the wind speed value is corresponding to the wind force level standard, for different wind force level standards, the influence of wind force on the driving stability of different vehicle types is judged, and the wind speed characteristics are quantified;

[0015] Based on the influence of temperature on vehicle parts and the correlation with road conditions, the temperature characteristics are quantified;

[0016] The visibility is classified according to different ranges to obtain the visibility range, the driver's operation under different visibility is observed through the monitoring data, the influence of the visibility range on the driver's operation is quantified, and the visibility characteristics are obtained;

[0017] Further, the quantified precipitation feature, wind speed feature, temperature feature and visibility feature are taken as the weather feature.

[0018] By using a geographic information system, the curvature radius of the curve at the accident location is measured, and whether there is a slope change at the curve is judged in combination with the terrain information in the geographic data; the satellite remote sensing image and the field three-dimensional modeling data are used to analyze the shielding condition of the view by the vegetation, buildings and terrain undulations around the curve, the visible range at different positions of the curve is simulated based on the driver's perspective, the proportion of the shielded view to the total view is calculated, and the curve feature is obtained;

[0019] Further, the climbing and downhill road sections in the road are identified, the starting point and the ending point of the climbing or downhill are determined according to the altitude change trend in the geographic data, the length and the average slope of the climbing and downhill are calculated, and the slope feature is obtained;

[0020] The minimum distance between the surrounding buildings and the road edge at the accident site is measured, and the surrounding building distribution feature is extracted;

[0021] Further, whether the accident location is at an intersection is identified according to the geographic data, if yes, the intersection type is further judged, the traffic signal, sign and marking at the intersection are viewed, and the road intersection feature is extracted;

[0022] Further, the curve feature, the slope feature, the surrounding building distribution feature and the road intersection feature are taken as the road environment feature.

[0023] The traffic sign is identified and located, the vehicle violation behavior is judged based on the traffic rules, and the violation deduction is performed; the Hough transform algorithm is used to identify the position and shape of the lane line, the lane line information is obtained by fusing and optimizing the lane line detection results of continuous multiple frames of images; the driving track of the vehicle on the road is tracked according to the lane line information, the center position, the driving direction and the relative position relationship with the lane line of the vehicle are recorded by using the target tracking algorithm, the violation behavior of the vehicle is judged based on the traffic rules, and the violation deduction is performed;

[0024] Further, when the meteorological data shows that the precipitation intensity reaches moderate rain and above, the use of the wiper by the driver is observed through the monitoring data to judge whether the rain causes the driver's vision to be blocked, and the violation deduction is performed; the driving stability of the vehicle is observed, if the vehicle slips on the waterlogged road and the driver does not take the measures of deceleration or avoidance, the violation deduction is performed; the driving behavior analysis in the strong wind weather is performed under the consideration of the operation in the poor visibility weather, and the violation deduction is performed;

[0025] Further, the violation deductions are accumulated to obtain the total violation risk score, and the traffic violation risk index is obtained by using linear change.

[0026] According to the precipitation and precipitation intensity information in the meteorological data, the interference degree of traffic is quantified; according to the wind speed value of the meteorological data, the interference intensity is judged according to the wind force level standard; according to the temperature in the meteorological data, the influence degree on the automobile parts is judged; the visibility is classified and quantified according to different ranges;

[0027] Based on the road environment feature, the interference is quantified, and the interference values of the curve features, slope features, surrounding building distribution features and road intersection features are evaluated;

[0028] Further, the interference values based on the weather features and the interference values based on the road environment features are added to obtain the total environmental interference intensity score; linear change is adopted to obtain the environmental interference intensity index.

[0029] For each identified traffic sign, through image feature extraction, it is judged whether the surface has damage, fading and deformation damage, and the damage score of the traffic sign is obtained; combined with geographic information system data and monitoring pictures, the source of the shelter around the traffic sign is determined, and the shelter score of the traffic sign is quantified by measuring the proportion of the shelter in the front field of view of the traffic sign; the damage score and the shelter score of the traffic sign are added to obtain the violation deduction of the traffic sign factor.

[0030] Based on the monitoring data, the target tracking algorithm is used to lock the pedestrians in the accident site and the surrounding area in real time, match the position of the pedestrians with the road environment features, judge whether the pedestrians are in a compliant position, and perform violation deduction; extract building distribution information, simulate the line of sight from the driver's perspective, analyze whether the building causes shelter to the pedestrians, and perform violation deduction;

[0031] Further, the violation deduction of the traffic sign factor, the violation deduction of the pedestrian position factor and the violation deduction of the surrounding building factor are added and linearly transformed to obtain the road facility correlation index.

[0032] Collect past traffic accident data from traffic police departments and insurance companies, use web crawler technology to collect traffic accident scene photos, videos and description information shared by the public from social media and traffic forums, supplement non-official records but with accident feature accident materials, and collect traffic accident data that can cover various accident cases in different regions, different time periods and various traffic scenes;

[0033] Further, a multi-layer perceptron model architecture is built, an input layer is designed, a hidden layer is configured, an output layer is set, a training set, a validation set and a test set are proportionally divided;The initial value of the weight adopts a random initialization method, and the weight and bias of the model are initialized;The traffic violation risk index, the environmental interference intensity index and the road facility correlation index in the training set are taken as the input, the actual proportion of human factors and natural factors of the traffic accident is taken as the target output, the model is input for forward propagation calculation;Neurons transmit information in turn according to weights and activation functions to obtain the predicted output of the model;The difference between the predicted output and the target output is used to calculate the gradient through the back propagation algorithm, and the weights and biases of each layer of the model are updated in reverse;After the model is trained for several rounds, the test set data is used to verify the model finally, and the evaluation index is used to measure the performance of the model;The model is optimized and adjusted.

[0034] Further, the threshold of the proportion of natural factors is set, when the proportion of natural factors exceeds the threshold, the reconstruction suggestion of the traffic accident location is provided based on the weather characteristics and the road environment characteristics, and the method is as follows:

[0035] For weather characteristics, when it is found in accident analysis that water accumulation caused by precipitation is the main natural factor, the drainage system of the accident location is checked and reconstructed, in urban roads, drainage outlets are set at a specified interval, and the size of the required drainage outlet is calculated according to the precipitation and the road area to ensure that the drainage capacity meets the requirements;For the water accumulation section of the curve and the downhill, the anti-skid pavement material is used for reconstruction;When the accident location is affected by the main natural factor of wind, it is suggested to set up windproof facilities and reinforce road facilities;When the temperature change of the accident location is the main natural factor, it is suggested to upgrade the road surface material and remind the driver to check the vehicle;When visibility is the main natural factor, it is suggested to improve the lighting facilities and set up the sight line guiding facilities;

[0036] For road environment characteristics, when the curve feature is the main natural factor, it is suggested to adjust the curve radius, set up the curve super-high and improve the sight line;When the slope feature is the main natural factor, it is suggested to adjust the slope and set up auxiliary facilities;When the surrounding building distribution feature is the main natural factor, it is suggested to adjust the building setback distance and control the building reflection;When the road intersection feature is the main natural factor, it is suggested to optimize the traffic control and reconstruct the channelization.

[0037] Compared with the prior art, the beneficial effects of the present application are:

[0038] 1、The present application collects and monitors meteorological and geographical data in all directions, and makes data interconnection through fine preprocessing and integration;When an accident occurs, the monitoring captures the details of the vehicle collision, and at the same time, the meteorological data synchronously feedbacks the accurate precipitation, wind speed and temperature at that time, and the geographical data provides the road slope, curve and surrounding building information.

[0039] 2、The application uses YOLOv5 analysis monitoring to accurately identify human-caused details; not only identifies vehicle violations such as lane changing, speeding, and not driving according to traffic signs, but also tracks driver eye movements and abnormal body movements; accurately locates the irregular path and speed of pedestrians, making it easier to determine responsibility.

[0040] 3、Through analysis and model training of past traffic accident data, the application can predict possible accident sections and potential risk factors in advance; combined with real-time collected data, real-time monitoring of the proportion of natural factors is performed, and when the proportion of natural factors is found to be close to or exceed the threshold, a timely warning is issued, and corresponding modification suggestions are provided. BRIEF DESCRIPTION OF DRAWINGS

[0041] Fig. 1 is a step schematic diagram of the safety warning signal sending control method applied to the intelligent monitoring device of the application;

[0042] Fig. 2 is a traffic accident occurrence site modification suggestion flowchart of the safety warning signal sending control method applied to the intelligent monitoring device of the application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the application will be described in detail below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0044] Embodiment: As shown in the figure, the application provides a technical solution, Figs. 1-2

[0045] According to one embodiment of the application, as shown in the figure, the safety warning signal sending control method applied to the intelligent monitoring device comprises the following steps: Fig. 1 As shown in the step schematic diagram of the safety warning signal sending control method applied to the intelligent monitoring device, the safety warning signal sending control method applied to the intelligent monitoring device comprises the following steps:

[0046] When a traffic accident occurs, the monitoring data, meteorological data and geographical data of the traffic accident occurrence site are acquired, data preprocessing and integration are performed, and traffic accident data is formed;

[0047] Based on the traffic accident data, the visual features of the target are extracted from the monitoring data using the YOLOv5 model, the target including vehicles, pedestrians and traffic signs; weather features are extracted according to the meteorological data, and precipitation, wind speed, temperature and visibility are quantified as characteristic values of the degree of influence on traffic; road environment features are extracted from the geographical data;​

[0048] Based on the traffic signs, whether the vehicle violates the rules stipulated by the traffic signs is identified, whether the driving trajectory of the vehicle conforms to the lane rules is judged, the operation score of the driver is obtained in combination with the weather characteristics, and a traffic violation risk index is formed; in combination with the weather characteristics and the road environment characteristics, an environmental interference intensity index is constituted; the damage situation of the traffic signs around the accident site is checked, the positions of the pedestrians and the surrounding building distribution are obtained, the shielding situation of the building to the pedestrians is checked, and a road facility correlation index is obtained;

[0049] Past traffic accident data is collected, a multi-layer perceptron model architecture is built, the multi-layer perceptron model is trained using the training set data, the weights and biases of the model are continuously adjusted through the back propagation algorithm, the traffic violation risk index, the environmental interference intensity index and the road facility correlation index are taken as inputs, and the actual proportion of human factors and natural factors of the traffic accident is taken as the target output;

[0050] A threshold of the proportion of natural factors is set, when the proportion of natural factors exceeds the threshold, a reconstruction suggestion of the traffic accident occurrence location is provided based on the weather characteristics and the road environment characteristics.

[0051] The YOLOv5 model is pre-trained on traffic scene image data, when a traffic accident occurs, the monitoring data in the traffic accident data is input into the YOLOv5 model in the form of an image sequence of one frame per second; for vehicles, the vehicle type, the vehicle body color and the license plate number are identified, and the driving trajectory of the vehicle before the accident is tracked; for pedestrians, the relative relationship between the position of the pedestrian at the time of the accident and the vehicle is judged; for traffic signs, the sign type and its position in the image are identified.

[0052] Further, based on the weather characteristics, the precipitation amount is divided into different magnitudes according to meteorological standards, the road environment characteristics are combined to estimate the water depth, and the working frequency of the wiper, the density of raindrops in front of the lens and the head movement of the driver are observed through the monitoring data to judge the degree of visual obstruction; based on the precipitation magnitude, the water depth and the degree of visual obstruction, the precipitation characteristics are quantified;

[0053] According to the wind speed value in the meteorological data, the wind speed value is corresponding to the wind force level standard, for different wind force level standards, the influence of wind force on the driving stability of different vehicle types is judged, and the wind speed characteristics are quantified;

[0054] Based on the influence of temperature on vehicle parts and the correlation with road conditions, the temperature characteristics are quantified;

[0055] The visibility is classified according to different ranges to obtain the visibility range, the driver's operation under different visibility is observed through the monitoring data, the influence of the visibility range on the driver's operation is quantified, and the visibility characteristics are obtained;

[0056] Further, the quantified precipitation feature, wind speed feature, temperature feature and visibility feature are taken as the weather feature.

[0057] By using the geographic information system, the curvature radius of the curve at the accident location is measured, and whether there is a slope change at the curve is judged in combination with the terrain information in the geographic data; according to the satellite remote sensing image and the field three-dimensional modeling data, the shielding condition of the vegetation, the building and the terrain undulation around the curve on the visual range is analyzed, the visible range at different positions of the curve is simulated based on the driver's visual angle, the proportion of the shielded visual range in the total visual range is calculated, and the curve feature is obtained;

[0058] Further, the climbing road section and the downhill road section in the road are identified, the starting point and the ending point of the climbing or downhill are determined according to the altitude change trend in the geographic data, the length and the average slope of the climbing and the downhill are calculated, and the slope feature is obtained;

[0059] The minimum distance between the surrounding buildings and the road edge is measured, and the surrounding building distribution feature is extracted;

[0060] Further, whether the accident location is at an intersection is identified according to the geographic data, if yes, the intersection type is further judged, the traffic signal, the sign and the marking at the intersection are viewed, and the road intersection feature is extracted;

[0061] Further, the curve feature, the slope feature, the surrounding building distribution feature and the road intersection feature are taken as the road environment feature.

[0062] In the embodiment, a two-way four-lane road located in the urban area is selected as the simulated accident occurring road section, the road section contains a curve with a curvature radius of about 60 meters and a slope of 3%, there are several buildings with a height of about 15 meters around the road section, the nearest distance between the buildings and the road edge is about 3 meters, and a traffic sign with a speed limit of 60 km / h is arranged at the exit of the curve. An experiment simulates a traffic accident occurring under the weather condition of moderate rain accompanied by 4-level wind and a visibility of about 800 meters.

[0063] A car is identified by the YOLOv5 model at the exit of a curve with a speed limit of 60 km / h. The vehicle's trajectory is analyzed within 100 meters before and after the sign using a target tracking algorithm. The car's speed reaches 75 km / h, which is 25% over the speed limit, resulting in a 5-point violation risk value. Additionally, the car did not use the turn signal before changing lanes, earning an additional 2 points. The monitoring image sequence is processed using the Hough transform algorithm to identify clear lane lines. After multiple frame fusion and optimization, stable lane line information is obtained. Another SUV is observed changing lanes three times within 10 seconds without using the turn signal, resulting in a total of 6 points for violation risk value. Under moderate rain, monitoring data shows that some vehicles' wipers are not functioning properly, affecting visibility. Three vehicles involved are each given 2 points. A van slips on a waterlogged road, and the driver fails to slow down in time, earning 3 points. Under strong winds, a truck deviates from its normal trajectory due to lateral wind, and the driver fails to adjust the direction, earning 2 points.

[0064] The total violation risk score is 24 points. Using linear transformation, a perfect score of 30 points corresponds to an index of 10. Therefore, the traffic violation risk index = 24 / 30 x 10 = 8.

[0065] The moderate rainfall causes some water accumulation on the road, with an estimated depth of 3-4 cm, assigned a value of 3. Rainwater significantly obstructs visibility, assigned a value of 3. Level 4 wind has some impact on the driving stability of high-body trucks, assigned a value of 2, and there is a risk of falling branches from surrounding trees, assigned a value of 1. The temperature of 20°C has no significant impact on vehicle components, assigned a value of 0. Visibility of 800 meters is mildly restricted, assigned a value of 1, but some drivers fail to turn on their fog lights as required, earning an additional 1 point. The weather feature interference value is 3+3+2+1+0+(1+1)=11. The curve radius of 60 meters is a medium curve, assigned a value of 2, and the slope of 3% is assigned an additional value of 1. The surrounding buildings are close to the road and pose a risk of obstructing visibility, assigned a value of 2. There are no special conditions at the road intersection, assigned a value of 0. The road environment feature interference value is 2+1+2+0=5.

[0066] The total environmental interference score is 11+5=16. After linear transformation, a perfect score of 20 points corresponds to an index of 10. Therefore, the environmental interference intensity index = 16 / 20 x 10 = 8.

[0067] The identified speed limit sign is checked, and image feature extraction shows that the sign has slight fading, earning 1 point. Combined with GIS data and monitoring images, it is found that the sign is obstructed by roadside branches by about 30% of the field of view, earning 2 points. The traffic sign factor violation deduction is a total of 3 points. Monitoring data tracks a pedestrian crossing the road at a non-pedestrian crossing, earning 2 points. The surrounding buildings' reflections at certain angles interfere with the driver's visibility, earning 1 point. The pedestrian position and building factor violation deduction is a total of 3 points.

[0068] The total violation deduction is 3+3=6 points. After linear transformation, the full score of 10 points corresponds to the index 0, so the road facility correlation index = 6 / 10*10 = 6.

[0069] Collect past traffic accident data from traffic police departments and insurance companies, use web crawler technology to collect photos, videos and description information of traffic accident scenes shared by the public from social media and traffic forums, supplement accident materials with accident characteristics that are not officially recorded, and collect traffic accident data that can cover various accident cases in different regions, different time periods and various traffic scenarios;

[0070] Further, a multi-layer perceptron model architecture is built, an input layer is designed, a hidden layer is configured, an output layer is set, and a training set, a validation set and a test set are proportionally divided;The initial value of the weight is randomly initialized, and the weight and bias of the model are initialized;The traffic violation risk index, the environmental interference intensity index and the road facility correlation index in the training set are used as input, and the actual proportion of human factors and natural factors of traffic accidents is used as target output, and the model is input for forward propagation calculation;Neurons transmit information in turn according to weights and activation functions to obtain the predicted output of the model;Using the difference between the predicted output and the target output, the gradient is calculated through the back propagation algorithm to update the weights and biases of each layer of the model;After the model is trained for several rounds, the test set data is used to verify the model finally, and the evaluation index is used to measure the performance of the model;Model optimization and adjustment are performed.

[0071] According to another embodiment of the present application, as Fig. 2 The traffic accident occurrence location modification suggestion flowchart of the safety warning signal sending control method applied to the intelligent monitoring device is shown, the threshold of the proportion of natural factors is set, when the proportion of natural factors exceeds the threshold, the modification suggestion of the traffic accident occurrence location is provided based on the weather characteristics and the road environment characteristics, and the method is as follows:

[0072] For weather characteristics, when it is found in accident analysis that water accumulation caused by precipitation is the main natural factor, the drainage system of the accident occurrence section is checked and modified, in urban roads, drainage outlets are set at a specified interval, and the size of the required drainage outlet is calculated according to the precipitation and the road area to ensure that the drainage capacity meets the requirements;For the water accumulation section of the curve and the downhill, anti-skid pavement materials are used for modification;When the accident site is affected by strong wind, it is suggested to set up windproof facilities and reinforce road facilities;When the temperature change of the accident site is the main natural factor, it is suggested to upgrade the road surface material and remind the driver to check the vehicle;When visibility is the main natural factor, it is suggested to improve the lighting facilities and set up the sight line guiding facilities;

[0073] For road environment features, when the bend feature is the main natural factor, it is recommended to adjust the bend radius, set the bend super-elevation, and improve the sight line; when the slope feature is the main natural factor, it is recommended to adjust the slope and set auxiliary facilities; when the surrounding building distribution feature is the main natural factor, it is recommended to adjust the building setback distance and control the building reflection; when the road intersection feature is the main natural factor, it is recommended to optimize the traffic control and reconstruct the channelization.

[0074] In this embodiment, a section of road located at the junction of a mountainous area and a city in a southern city is selected as the experimental object. The total length of the road is about 2 kilometers, including a 500-meter-long continuous bend, with the minimum bend radius of 60 meters and a slope of 4%; there are some old buildings around the road, with the nearest distance of about 5 meters from the road; at one intersection of the road, the traffic flow is large.

[0075] The traffic accident data of this section of road for the past 3 years is collected, a total of 50 accidents are counted. After analysis, it is found that 20 of the accidents have obvious natural factors, including precipitation, strong wind, low visibility and other weather factors, as well as road bends, slopes and other environmental factors. According to the historical data and expert evaluation, the threshold of the proportion of natural factors is set to 40%.

[0076] After a heavy rain, a three-car rear-end collision accident occurred on this section of road. At that time, the precipitation reached 50 mm / hour, and the continuous rainfall lasted for 2 hours. Accurate precipitation data was obtained through meteorological monitoring stations; the process of the accident was recorded by road monitoring video; the slope and bend of the road were obtained through geographic information system (GIS). After analysis, the proportion of natural factors in this accident reached 60%, mainly due to heavy rain causing serious road waterlogging, and the vehicle lost control when passing through the bend.

[0077] The drainage system was inspected and reconstructed, and according to the standard of setting a drainage port every 30 meters, 15 drainage ports were added on this section of road. At the same time, according to the precipitation and road area calculation, the size of the drainage port was expanded from the original diameter of 30 cm to 50 cm. For the bend and downhill waterlogging sections, asphalt pavement materials with better anti-skid performance were used for re-paving.

[0078] Under a 6-level strong wind, a high vehicle collided with a small car when it was driving on this section of road due to lateral wind deviation from the normal driving track. The wind speed recorded by meteorological data was 12 m / s; the accident process was recorded by monitoring video; the geographic data showed that the accident occurred on a relatively open section of road without effective wind protection facilities. After calculation, the proportion of natural factors in this accident was 50%.

[0079] Two rows of tall poplar trees were planted on both sides of the road as windbreak, with a spacing of 4 meters and a row spacing of 5 meters. The facilities around the road, such as street lamps and billboards, were reinforced by adding fixed supports and wind-resistant cables.

[0080] A car ran off the road when it was passing the curve of the section due to excessive speed, causing vehicle damage and personal injury. The vehicle trajectory and speed were obtained through the monitoring video; the curvature radius and slope of the curve were measured by GIS data. The analysis showed that the natural factors accounted for 45% of the accident, mainly due to the small radius and large slope of the curve.

[0081] The curve was reconstructed by expanding the radius from 60 meters to 80 meters through excavation of part of the mountain and filling. An extra-high curve was set up with a slope of 3%, making the vehicle more stable when driving on the curve. Some trees that blocked the view were removed around the curve, and wide-angle mirrors were installed to improve the driver's view.

[0082] It is apparent to those skilled in the art that the application is not limited to the details of the foregoing exemplary embodiments and that the application can be implemented in other particular forms without departing from the spirit or essential characteristics of the application. The presently disclosed embodiments are therefore considered in all respects to be illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalents are intended to be embraced therein. No feature of the application is to be construed as limiting the scope of the claims to their precise form.

Claims

1. A security early warning signal sending control method applied to an intelligent monitoring device, characterized in that, The method comprises the following steps: When a traffic accident occurs, obtain monitoring data, meteorological data and geographical data of the traffic accident occurrence site, perform data preprocessing and integration, and form traffic accident data; Based on the traffic accident data, the visual features of the target are extracted from the monitoring data using the YOLOv5 model, the target including a vehicle, a pedestrian and a traffic sign; the weather features are extracted according to the meteorological data, and the precipitation, wind speed, temperature and visibility are quantified as characteristic values of the influence degree on the traffic; the road environment features are extracted from the geographical data; Based on the traffic sign, whether the vehicle violates the rules stipulated by the traffic sign is identified, whether the driving track of the vehicle conforms to the lane rules is judged, the operation score of the driver is obtained in combination with the weather features, and a traffic violation risk index is formed; in combination with the weather features and the road environment features, an environmental interference intensity index is constituted; the damage of the traffic sign around the accident site is checked, the position of the pedestrian and the distribution of the surrounding buildings are obtained, and the shielding of the buildings to the pedestrian is checked to obtain a road facility correlation index; The traffic sign is identified and located, the vehicle violation behavior is judged based on the traffic rules, and the violation score is deducted; the position and shape of the lane line are identified by using the Hough transform algorithm, the lane line information is obtained by fusing and optimizing the lane line detection results of continuous multiple frames of images, the driving track of the vehicle on the road is tracked according to the lane line information, the center position, driving direction and relative position relationship with the lane line of the vehicle are recorded by using the target tracking algorithm, the violation behavior of the vehicle is judged based on the traffic rules, and the violation score is deducted; When the meteorological data shows that the precipitation intensity reaches the level of moderate rain and above, the use of the windshield wiper by the driver is observed through the monitoring data, whether the rain causes the driver's vision to be blocked is judged, and the violation score is deducted; the driving stability of the vehicle is observed, if the vehicle slips on the waterlogged road and the driver does not take measures to slow down or avoid, the violation score is deducted; the operation in the weather with insufficient visibility is considered, the driving behavior in the strong wind weather is analyzed, and the violation score is deducted; The violation scores are added up to obtain a total violation risk score, and a traffic violation risk index is obtained by using linear change; The interference degree on the traffic is quantified according to the precipitation amount and precipitation intensity information in the meteorological data; the interference intensity is judged according to the wind speed value of the meteorological data; the influence degree on the automobile parts is judged according to the temperature in the meteorological data; the visibility is quantified by grading according to different ranges; The interference values based on the weather features and the interference values based on the road environment features are added up to obtain a total environmental interference intensity score; an environmental interference intensity index is obtained by using linear change; ​ Collect past traffic accident data, build a multi-layer perceptron model architecture, train the multi-layer perceptron model using the training set data, and continuously adjust the weights and biases of the model through the back propagation algorithm. The traffic violation risk index, environmental interference intensity index, and road facility correlation index are used as inputs, and the actual proportion of human factors and natural factors in traffic accidents is used as the target output. Set a threshold for the proportion of natural factors. When the proportion of natural factors exceeds the threshold, provide reconstruction suggestions for the location of the traffic accident based on weather characteristics and road environment characteristics.

2. The security early warning signal sending control method applied to the intelligent monitoring device according to claim 1, characterized in that: Pre-train the YOLOv5 model on traffic scene image data. When a traffic accident occurs, input the monitoring data in the traffic accident data into the YOLOv5 model in image sequences at a rate of one frame per second. For vehicles, identify the vehicle type, body color, and license plate number, and track the vehicle's driving trajectory before the accident. For pedestrians, determine the relative position of the pedestrian to the vehicle at the time of the accident. For traffic signs, identify the sign type and its position in the image.

3. The security early warning signal sending control method applied to the intelligent monitoring device according to claim 2, characterized in that: Based on the weather characteristics, divide the precipitation amount into different levels according to meteorological standards, and estimate the water depth based on the road environment characteristics. Observe the rain wiper working frequency, the density of raindrops in front of the lens, and the degree of visual obstruction by the driver's head movement through monitoring data. Quantify the precipitation characteristics based on the precipitation level, water depth, and visual obstruction degree. According to the wind speed value in the meteorological data, corresponding to the wind force level standard, for different wind force level standards, for different vehicle types, judge the influence of wind force on vehicle driving stability, and quantify the wind speed characteristics. Quantify the temperature characteristics based on the influence of temperature on vehicle components and the correlation with road conditions. Classify visibility into different ranges to obtain the visibility range. Observe the driver's operation under different visibility through monitoring data, quantify the impact of visibility range on driver operation, and obtain the visibility characteristics. Use the quantified precipitation characteristics, wind speed characteristics, temperature characteristics, and visibility characteristics as weather characteristics.

4. The security early warning signal sending control method applied to the intelligent monitoring device according to claim 3, characterized in that: Use geographic information systems to measure the curvature radius of the curve at the accident location, and determine whether there is a slope change at the curve based on the terrain information in the geographic data. Analyze the obstruction of the view by vegetation, buildings, and terrain undulations around the curve based on satellite remote sensing images and field three-dimensional modeling data. Simulate the visible range at different positions on the curve from the driver's perspective, calculate the proportion of the obstructed view to the total view, and obtain the curve characteristics. Identify the climbing and descending road sections in the road, determine the starting and ending points of climbing or descending based on the altitude change trend in the geographic data, calculate the length and average slope of the climbing and descending, and obtain the slope characteristics. Measure the minimum distance between the surrounding buildings and the road edge, and extract the surrounding building distribution characteristics. Identify whether the accident location is at an intersection based on geographic data. If so, further determine the type of intersection, view the traffic signals, signs, and markings at the intersection, and extract the road intersection characteristics. The curve characteristics, slope characteristics, surrounding building distribution characteristics, and road intersection characteristics are used as road environment characteristics.

5. The security warning signal transmission control method for intelligent monitoring devices according to claim 1, characterized in that: For each identified traffic sign, through image feature extraction, determine whether there is damage such as damage, fading and deformation on the surface, and obtain the damage score of the traffic sign; combined with geographic information system data and monitoring screen, determine the source of the shelter around the traffic sign, and measure the proportion of the shelter in the field of view in front of the traffic sign to quantify the shelter score of the traffic sign; add the damage score and the shelter score of the traffic sign to obtain the violation deduction of the traffic sign factor.

6. The security early warning signal sending control method applied to the intelligent monitoring device according to claim 5, characterized in that: Based on the monitoring data, the target tracking algorithm is used to lock the pedestrians in the accident site and the surrounding area in real time, match the position of the pedestrians with the road environment characteristics, judge whether the pedestrians are in a compliant position, and deduct the violation score; Extract the building distribution information, simulate the line of sight from the driver's perspective, analyze whether the building causes shelter to the pedestrians, and deduct the violation score; Add the violation deduction of the traffic sign factor, the violation deduction of the pedestrian position factor and the violation deduction of the surrounding building factor, perform linear transformation, and obtain the road facility correlation index.

7. The security early warning signal sending control method applied to the intelligent monitoring device according to claim 1, characterized in that: Collect past traffic accident data from traffic police departments and insurance companies, use web crawler technology to collect traffic accident scene photos, videos and description information shared by the public from social media and traffic forums, supplement accident materials with accident characteristics but not official records, and collect traffic accident data that can cover various accident cases in different regions, different time periods and various traffic scenes; Build a multi-layer perceptron model architecture, design the input layer, configure the hidden layer, set the output layer, and divide the training set, validation set and test set in proportion; the initial value of the weight is randomly initialized, and the weight and bias of the model are initialized; the traffic violation risk index, environmental interference intensity index and road facility correlation index in the training set are used as input, the actual proportion of human factors and natural factors of traffic accidents is used as target output, and the model is input for forward propagation calculation; neurons transmit information in turn according to weights and activation functions to obtain the predicted output of the model; use the difference between the predicted output and the target output to calculate the gradient through the back propagation algorithm, and update the weights and biases of each layer of the model in reverse; after the model is trained for several rounds, use the test set data to verify the model finally, and use evaluation indicators to measure the performance of the model; optimize and adjust the model.

8. The security early warning signal sending control method applied to the intelligent monitoring device according to claim 7, characterized in that: Set a threshold for the proportion of natural factors. When the proportion of natural factors exceeds the threshold, provide reconstruction suggestions for the location of the traffic accident based on weather characteristics and road environment characteristics, as follows: For weather characteristics, when precipitation-induced water accumulation is the main natural factor found in accident analysis, check and reconstruct the drainage system of the accident-occurring section. In urban roads, set drainage outlets at specified intervals and calculate the size of the required drainage outlets according to the amount of precipitation and road area to ensure that the drainage capacity meets the requirements. For sections with water accumulation on curves and slopes, use anti-skid pavement materials for reconstruction. When strong winds affect the accident site, it is recommended to set up windproof facilities and reinforce road facilities. When temperature changes at the accident site are the main natural factor, it is recommended to upgrade the road surface material and provide vehicle inspection prompts for drivers. When visibility is the main natural factor, it is recommended to improve lighting facilities and set up sight line guiding facilities. For road environment characteristics, when curve characteristics are the main natural factor, it is recommended to adjust the curve radius, set up curve super-elevation, and improve the sight line. When slope characteristics are the main natural factor, it is recommended to adjust the slope and set up auxiliary facilities. When the distribution characteristics of surrounding buildings are the main natural factor, it is recommended to adjust the setback distance of buildings and control the reflection of buildings. When road intersection characteristics are the main natural factor, it is recommended to optimize traffic control and reconstruct the channel.

Images