Road site security monitoring system and method based on big data and artificial intelligence

Abstract

The present invention relates to the technical field of road traffic management, and in particular to a road on-site security monitoring system and method based on big data and artificial intelligence. The method comprises: collecting vehicle data entering a target monitoring area according to a preset collection frequency, and using the vehicle data to monitor vehicle operation characteristics and driver status characteristics; predicting the vehicle intrusion probability through a pre-trained danger warning model based on the vehicle operation characteristics and driver status characteristics, and generating corresponding alarm information; determining the vehicle intrusion mode, filtering the alarm information, and extracting effective alarm information; and after obtaining the effective alarm information, constructing a multimodal hierarchical alarm strategy, providing real-time feedback of target monitoring area image data, and generating an optimal risk avoidance path.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic road construction and maintenance, and in particular to a road site security monitoring system and method based on big data and artificial intelligence. Background Art

[0002] Highway traffic management is a complex project involving multiple fields and departments. Its purpose is to ensure the safe, smooth and efficient operation of highways, including road maintenance and construction management.

[0003] Road maintenance and construction management requires that during maintenance and construction, construction warning areas, buffer zones, and work zones must be set up in accordance with standards to alert vehicles and prevent them from intruding into the work zone, thereby ensuring the safety of personnel in the work area.

[0004] With the development of intelligent technologies, road maintenance and construction management are also moving towards intelligentization. For example, by collecting vehicle operation data from warning zones and buffer zones upstream of the work area, including surveillance video data, speed radar data, and driver image data, and using artificial intelligence models to integrate and analyze the corresponding data, dangers (vehicle intrusions) can be predicted, enabling visual management and early warning of the work area and the areas upstream of the work area.

[0005] Traditional technologies have problems such as single early warning means, inability to link between devices and between devices and monitoring centers, and low early warning efficiency. Summary of the Invention

[0006] The present invention collects a variety of data, monitors the vehicle's operating status, predicts the probability of a vehicle intruding into a work area through a danger warning model, and performs graded warnings to improve the warning effect.

[0007] The technical solution proposed by the present invention is: a road on-site security monitoring method based on big data and artificial intelligence, the method comprising:

[0008] Collecting vehicle data entering the target monitoring area at a preset collection frequency, and using the vehicle data to monitor vehicle operation characteristics and driver status characteristics;

[0009] Based on the vehicle operation characteristics and driver status characteristics, the pre-trained danger warning model is used to predict the vehicle intrusion probability and generate corresponding alarm information;

[0010] Determine vehicle intrusion patterns, filter alarm information, and extract effective alarm information;

[0011] After obtaining effective alarm information, a multimodal hierarchical alarm strategy is constructed to provide real-time feedback of target monitoring area image data and generate the optimal risk avoidance path.

[0012] Preferably, the vehicle data includes video data of vehicles running in the target monitoring area, infrared image data of drivers, vehicle speed data, vehicle identity data and passing time data;

[0013] The monitoring of vehicle operation characteristics and driver status characteristics using the vehicle data includes:

[0014] Align the video data and speed data of vehicles running in the target monitoring area on the time axis;

[0015] Extract the vehicle's speed characteristics from the vehicle's speed data, including the vehicle's instantaneous speed and acceleration ;

[0016] Extract vehicle direction change features, including heading angle deviation, from vehicle video data , lateral displacement rate and trajectory curvature ;

[0017] in, , Indicates the direction of vehicle travel. Indicates lane direction;

[0018] , , represents the vehicle displacement increment;

[0019] Extracting driver status features, including distraction indicators, from driver infrared image data , head offset angle , blink frequency ;in, Indicates how long the eyes are looking at the road. Indicates the total fixation duration.

[0020] Preferably, the method of predicting the vehicle intrusion probability based on the vehicle operation characteristics and the driver status characteristics through a pre-trained danger warning model and generating corresponding alarm information includes:

[0021] After normalizing the speed feature, direction change feature and driver state feature, the initial vehicle feature vector is constructed ;in They represent the normalized instantaneous velocity, acceleration, heading angle deviation, lateral displacement rate, trajectory curvature, Euclidean distance from the vehicle to the warning area boundary, and attention distraction index respectively;

[0022] Construct a feature vector sequence within the acquisition time window ; Indicates the length of the acquisition time window; Indicates the end point of the collection period;

[0023] Inputting the feature vector into a pre-trained risk prediction model, wherein the risk prediction model is a logistic regression model;

[0024] Predicting vehicle intrusion probability through danger prediction model ,in, , represents the intercept, Indicates the Time point The weight of each feature; Indicates time No. Features Indicates the number of eigenvector elements;

[0025] If the invasion probability When the value is greater than the preset alarm threshold, an alarm message is generated to remind on-site staff that there is a vehicle intrusion;

[0026] The image data of the corresponding vehicle is obtained, and the image data of the vehicle is associated with the alarm information and sent to the cloud to remind the monitoring personnel that there is a vehicle intrusion.

[0027] Preferably, the method of predicting the vehicle intrusion probability based on the vehicle operation characteristics and the driver status characteristics through a pre-trained danger warning model and generating corresponding alarm information also includes:

[0028] Obtain the motion characteristics and driver status characteristics of vehicles in adjacent lanes to the target monitoring area and predict vehicle lane change intrusion, including:

[0029] Acquire running image data of vehicles in adjacent lanes, vehicle running speed and driver status image data;

[0030] Align vehicle operation image data, vehicle operation speed, and driver status image data through hardware-triggered timestamp synchronization;

[0031] Extract vehicle motion features in adjacent lanes, including lane line angles and the vehicle's relative position change rate ,in, Indicates the distance the vehicle moves along the lane line during the collection interval. Indicates the distance the vehicle moves in the direction perpendicular to the lane line during the collection interval;

[0032] After normalizing the motion features of vehicles in adjacent lanes, a fusion vehicle feature vector is constructed. ;in, Represent the normalized lane line angle and vehicle relative position change rate respectively;

[0033] Construct a fusion feature vector sequence within the acquisition time window ;

[0034] The vehicle extended feature vector is input into the pre-trained enhanced intrusion risk prediction model, and the fused intrusion probability is output. ; Specifically include:

[0035] An enhanced intrusion risk prediction model is constructed through a bidirectional LSTM network, wherein the enhanced intrusion risk prediction model includes a motion branch, a driver branch, and a decision fusion branch;

[0036] Output motion feature vector through motion branch , output the driver feature vector through the driver branch ;

[0037] Fusion invasion probability ;in, represents the output layer fusion matrix, represents the output layer bias; represents the motion feature fusion weight and driver feature fusion weight;

[0038] If the fused intrusion probability is greater than the preset alarm threshold, an alarm message is output.

[0039] Preferably, the method of predicting the vehicle intrusion probability based on the vehicle operation characteristics and the driver status characteristics through a pre-trained danger warning model and generating corresponding alarm information also includes:

[0040] Divide the target monitoring area into multiple response segments, assign warning weights to each response segment, and optimize the adaptability of the alarm strategy, including:

[0041] The target monitoring area is divided into corresponding segments, each corresponding segment ;in, Respectively represent the target monitoring area position and the target monitoring area end position; ;

[0042] Assign a base weight to each segment ;

[0043] Get traffic density factor ;

[0044] Get visibility factor ;in, Indicates visibility;

[0045] Obtain historical accident rates ;in, represents the number of vehicles in each response segment, Indicates the density of labeled vehicles, Indicates the number of historical accidents in each response segment obtained from the database, It represents the total number of historical accidents in the target monitoring area obtained from the database;

[0046] The dynamic weight of each response segment ;in, They represent the traffic density sensitivity coefficient, the minimum time for a vehicle to reach the work area from its response section, the actual time for a vehicle to reach the work area from its response section, the visibility attenuation coefficient, and the accident intensification coefficient respectively;

[0047] Develop a segmented alarm strategy, including:

[0048] Divide alarm intensity levels according to dynamic weights ;

[0049] According to different alarm intensity levels, single-mode graded alarm is carried out, including:

[0050] if , then a single sound alarm will sound at the end point of the corresponding response segment;

[0051] if , then an intermittent sound alarm will be issued at the end point of the corresponding response segment;

[0052] if , a high-frequency sound alarm will be issued at the end point of the corresponding response segment.

[0053] Preferably, the determining of vehicle intrusion mode, filtering of alarm information, and extraction of effective alarm information include:

[0054] Distinguish intrusion patterns in the lane and adjacent lanes within the target monitoring area, monitor vehicle exit behavior in real time, and optimize intrusion probability prediction, including:

[0055] Perform lane classification and intrusion type determination, including:

[0056] Encode lane space, lane coding ;

[0057] if , then it is judged that the vehicle intrudes straight into the lane;

[0058] if , then it is judged that the adjacent lane is intruded;

[0059] if , it is judged that the vehicle is out of control and intrudes;

[0060] Monitor the vehicle's operating status and determine whether the vehicle has changed out of the corresponding motion state, including:

[0061] Set the recovery judgment indicator ;in, Indicates that the vehicle moves laterally in the original lane. Indicates that the vehicle is actively driving to avoid a maneuver;

[0062] Calculate the corrected fusion invasion probability ,in, represents the probability attenuation coefficient; time interval;

[0063] If the fusion intrusion probability is greater than the preset alarm threshold and When the vehicle changes its intrusion state, the alarm information is considered invalid and the alarm information output is stopped;

[0064] If the fusion intrusion probability is greater than the preset alarm threshold and , it is determined that the vehicle has not changed the intrusion status, the alarm information is a valid alarm, and the alarm information is kept output.

[0065] Preferably, the multi-modal hierarchical alarm strategy is constructed to provide real-time feedback of target monitoring area image data, including:

[0066] A multi-modal hierarchical alarm strategy is constructed based on the alarm intensity level and recovery judgment indicators, including:

[0067] Obtain alarm intensity level, recovery judgment index, video stream of target monitoring area and video stream of working area;

[0068] if and , sound and light alarm is performed in the last response segment of the target monitoring area;

[0069] if and , sound and light alarms are issued in the working area, and alarm information and video streams of the target monitoring area are sent to the monitoring center, and intruding vehicles are marked in the video streams of the target monitoring area;

[0070] if and If the time interval is greater than 1.5 seconds, an audible and visual alarm will be sounded in the working area, and a rescue request will be sent to the monitoring center;

[0071] Generate the optimal risk avoidance path and feed it back to the monitoring center;

[0072] Generating the optimal risk avoidance path includes:

[0073] Obtain image information of the work area and the coordinate information of the staff from the GIS system. Combined with the predicted trajectory of the intruding vehicle, the optimal avoidance path is calculated, including:

[0074] Predicting intruding vehicle trajectories:

[0075] ;in, Represent the vehicle coordinates at the current moment, represents the prediction time interval, Indicates the acquisition frequency;

[0076] Predict vehicle driving direction angle ;

[0077] Obtain staff coordinate information and construct threat areas ;

[0078] ,in, Indicates the radius of the safe area; Indicates any coordinate point within the threat area;

[0079] Combine the vehicle's predicted direction angle to construct the optimal risk avoidance path:

[0080] ,in, They represent the direction vector of the staff and the direction vector of the nearest safe haven respectively. Indicates the normalized value of the direction angle from the worker to the shelter; Represents the normalized value of the vehicle's predicted driving direction angle; Represent position weight and angle weight respectively; Indicates the coordinates of the nearest safe haven;

[0081] get and Finally, an alternative escape path is generated through the Bezier curve, including:

[0082] Direct safe-haven path ,in, represents the correction factor, ;

[0083] Left avoidance path , where the lateral offset ;

[0084] Right avoidance path ;

[0085] Calculation acquisition 、 、 、 , generate alarm instructions and send them to the staff in the work area to guide them to take emergency precautions.

[0086] Preferably, the method further includes dynamically adjusting the warning state according to the vehicle intrusion probability, the recovery behavior, and the vehicle location, allocating computing resources and communication priority according to the warning state, and reducing the warning delay, including:

[0087] Obtaining the status of each response segment, wherein the status of the response segment includes monitoring state, warning state, emergency state and mitigation state;

[0088] when When , the corresponding response segment switches from monitoring state to warning state; Indicates the distance between the vehicle and the working area; Indicates a high risk threshold;

[0089] when When , the corresponding response segment switches from the warning state to the mitigation state;

[0090] when When the corresponding response segment is converted from the warning state to the emergency state;

[0091] when When the corresponding response stage changes from mitigation state to monitoring state, the low risk threshold .

[0092] A road site security monitoring system based on big data and artificial intelligence, wherein the system is used to execute the road site security monitoring method based on big data and artificial intelligence.

[0093] A computer-readable storage medium stores a computer program, which is executed by a processor to implement the above-mentioned road site security monitoring method based on big data and artificial intelligence.

[0094] Beneficial effects of the present invention:

[0095] 1. The present invention predicts the probability of a vehicle intruding into a work area through a hazard warning model, and screens effective alarm information based on the monitoring of the intrusion pattern and the vehicle's recovery behavior. Based on the effective alarm information, the present invention adopts corresponding alarm strategies (single-modal graded alarm and multi-modal graded alarm), and generates corresponding evasion paths to guide workers to avoid danger.

[0096] 2. In the present invention, the target monitoring area is divided into multiple response segments for segmented monitoring and early warning, and vehicles are reminded in advance. In the working area, sound and light alarms are used to remind vehicles to slow down and workers to avoid. The image data of the working area, upstream warning area and buffer area are transmitted to the monitoring center in real time, which facilitates visual management of the site. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 This is a flow chart of the road site security monitoring method based on big data and artificial intelligence of the present invention. DETAILED DESCRIPTION

[0098] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0099] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0100] Example 1:

[0101] refer to Figure 1 The technical solution provided by the present invention is: a road site security monitoring method based on big data and artificial intelligence, comprising the following steps:

[0102] Step 1: Collect vehicle data entering the target monitoring area according to a preset collection frequency; the vehicle data includes video data of vehicles running in the target monitoring area, infrared image data of drivers, vehicle speed data, vehicle identity data and passing time data;

[0103] Utilizing the vehicle data to monitor vehicle operation characteristics and driver status characteristics; specifically including:

[0104] Align the video data and speed data of vehicles running in the target monitoring area on the time axis;

[0105] Extract the vehicle's speed characteristics from the vehicle's speed data, including the vehicle's instantaneous speed and acceleration ;

[0106] Extract vehicle direction change features, including heading angle deviation, from vehicle video data , lateral displacement rate and trajectory curvature ;

[0107] in, , Indicates the direction of vehicle travel. Indicates lane direction;

[0108] , , represents the vehicle displacement increment;

[0109] Extracting driver status features, including distraction indicators, from driver infrared image data , head offset angle , blink frequency ;in, Indicates how long the eyes are looking at the road. represents the total fixation duration;

[0110] Step 2: Based on the vehicle operation characteristics and driver status characteristics, the pre-trained danger warning model is used to predict the vehicle intrusion probability and generate corresponding alarm information. Specifically, it includes:

[0111] After normalizing the speed feature, direction change feature and driver state feature, the initial vehicle feature vector is constructed ;in They represent the normalized instantaneous velocity, acceleration, heading angle deviation, lateral displacement rate, trajectory curvature, Euclidean distance from the vehicle to the warning area boundary, and attention distraction index respectively;

[0112] Construct a feature vector sequence within the acquisition time window ; Indicates the length of the acquisition time window; Indicates the end point of the collection period;

[0113] Inputting the feature vector into a pre-trained risk prediction model, wherein the risk prediction model is a logistic regression model;

[0114] Predicting vehicle intrusion probability through danger prediction model ,in, , represents the intercept, Indicates the Time point The weight of each feature; Indicates time No. Features Indicates the number of eigenvector elements;

[0115] If the invasion probability When the value is greater than the preset alarm threshold, an alarm message is generated to remind on-site staff that there is a vehicle intrusion;

[0116] The image data of the corresponding vehicle is obtained, and the image data of the vehicle is associated with the alarm information and sent to the cloud to remind the monitoring personnel that there is a vehicle intrusion.

[0117] Step 3: Determine the vehicle intrusion mode, filter the alarm information, and extract the effective alarm information. This includes the following steps:

[0118] Distinguish intrusion patterns in the lane and adjacent lanes within the target monitoring area, monitor vehicle exit behavior in real time, and optimize intrusion probability prediction, including:

[0119] Perform lane classification and intrusion type determination, including:

[0120] Encode lane space, lane coding ;

[0121] if , then it is judged that the vehicle intrudes straight into the lane;

[0122] if , then it is judged that the adjacent lane is intruded;

[0123] if , it is judged that the vehicle is out of control and intrudes;

[0124] Monitor the vehicle's operating status and determine whether the vehicle has changed out of the corresponding motion state, including:

[0125] Set the recovery judgment indicator ;in, Indicates that the vehicle moves laterally in the original lane. Indicates that the vehicle is actively driving to avoid a maneuver;

[0126] Calculate the corrected fusion invasion probability ,in, represents the probability attenuation coefficient; time interval;

[0127] If the fusion intrusion probability is greater than the preset alarm threshold and When the vehicle changes its intrusion state, the alarm information is considered invalid and the alarm information output is stopped;

[0128] If the fusion intrusion probability is greater than the preset alarm threshold and , it is determined that the vehicle has not changed the intrusion status, the alarm information is a valid alarm, and the alarm information is kept output.

[0129] Step 4: After obtaining effective alarm information, a multimodal hierarchical alarm strategy is constructed to provide real-time feedback of target monitoring area image data and generate the optimal risk avoidance path.

[0130] Among them, building a multi-modal hierarchical alarm strategy and real-time feedback of target monitoring area image data includes the following steps:

[0131] A multi-modal hierarchical alarm strategy is constructed based on the alarm intensity level and recovery judgment indicators, specifically:

[0132] Obtain alarm intensity level, recovery judgment index, video stream of target monitoring area and video stream of working area;

[0133] if and , sound and light alarms are issued in the last response section of the target monitoring area; for example, the sound and light alarm device (maintenance sentry) reminds intruding vehicles to slow down, and the LED display device at the rear reminds the staff in the working area;

[0134] if and , sound and light alarms are issued in the working area, and alarm information and video streams of the target monitoring area are sent to the monitoring center, and intruding vehicles are marked in the video streams of the target monitoring area;

[0135] if and If the time interval is greater than 1.5 seconds, an audible and visual alarm will be sounded in the working area, and a rescue request will be sent to the monitoring center;

[0136] Generate the optimal avoidance path and feed the optimal avoidance path and work area video stream back to the monitoring center;

[0137] Generating the optimal risk avoidance path includes:

[0138] Obtain image information of the work area and the coordinate information of the staff from the GIS system. Combined with the predicted trajectory of the intruding vehicle, the optimal avoidance path is calculated, including:

[0139] Predicting intruding vehicle trajectories:

[0140] ;in, Represent the vehicle coordinates at the current moment, represents the prediction time interval, Indicates the acquisition frequency;

[0141] Predict vehicle driving direction angle ;

[0142] Obtain staff coordinate information and construct threat areas ;

[0143] ,in, Indicates the radius of the safe area; Indicates any coordinate point within the threat area;

[0144] Combine the vehicle's predicted direction angle to construct the optimal risk avoidance path:

[0145] ,in, They represent the direction vector of the staff and the direction vector of the nearest safe haven respectively. Indicates the normalized value of the direction angle from the worker to the shelter; Represents the normalized value of the vehicle's predicted driving direction angle; Represent position weight and angle weight respectively; Indicates the coordinates of the nearest safe haven.

[0146] get and Finally, an alternative escape path is generated through the Bezier curve, including:

[0147] Direct safe-haven path ,in, represents the correction factor, ; Based on the safety area radius; if the safety radius is less than 2 meters, then , if the safety radius is greater than 2 meters, then ;

[0148] Left avoidance path , where the lateral offset ;

[0149] Right avoidance path ;

[0150] Calculation acquisition 、 、 、 , generate alarm instructions and send them to the staff in the work area to guide them to take emergency precautions.

[0151] For example, in this embodiment, a wearable device (smart helmet) receives and plays an alarm command (e.g., move 5 meters to the right or 3 meters back). Video data from the alarm area and buffer area is transmitted to a monitoring center for remote monitoring.

[0152] Example 2:

[0153] On roads with multiple lanes, vehicle intrusions can occur from either the vehicle's own lane (where the vehicle's lane coincides with the lane in the road construction area) or from adjacent lanes. Therefore, when predicting intrusion risk, it's necessary to consider the vehicle's intrusion pattern. To this end, based on Example 1, we enhance the ability to predict lane change intrusions by leveraging the motion characteristics of vehicles in adjacent lanes and the driver's state. The specific solution is as follows:

[0154] Obtain the motion characteristics and driver status characteristics of vehicles in adjacent lanes to the target monitoring area and predict vehicle lane change intrusion, including:

[0155] Acquire vehicle movement image data, vehicle speed, and driver status image data from adjacent lanes. In this embodiment, high-definition cameras deployed on one side of the lane capture vehicle movement images from adjacent lanes; speed measurement equipment (maintenance sentinels or radar anti-intrusion equipment) measures vehicle speed; and infrared cameras (with face tracking) capture infrared images of the driver's status.

[0156] Align vehicle operation image data, vehicle operation speed, and driver status image data through hardware-triggered timestamp synchronization;

[0157] Extract vehicle motion features in adjacent lanes, including lane line angles and the vehicle's relative position change rate ,in, Indicates the distance the vehicle moves along the lane line during the collection interval. Indicates the distance the vehicle moves in the direction perpendicular to the lane line during the collection interval;

[0158] After normalizing the motion features of vehicles in adjacent lanes, a fusion vehicle feature vector is constructed. ;in, Represent the normalized lane line angle and vehicle relative position change rate respectively;

[0159] Construct a fusion feature vector sequence within the acquisition time window .

[0160] The vehicle extended feature vector is input into the pre-trained enhanced intrusion risk prediction model, and the fused intrusion probability is output. ; Specifically include:

[0161] An enhanced intrusion risk prediction model is constructed through a bidirectional LSTM network, wherein the enhanced intrusion risk prediction model includes a motion branch, a driver branch, and a decision fusion branch;

[0162] Output motion feature vector through motion branch , output the driver feature vector through the driver branch ;

[0163] Fusion invasion probability ;in, represents the output layer fusion matrix, represents the output layer bias; represents the motion feature fusion weight and driver feature fusion weight;

[0164] If the fused intrusion probability is greater than the preset alarm threshold, an alarm message is output.

[0165] Example 3:

[0166] During road construction, an alarm zone and a buffer zone are set up upstream of the work area. In this embodiment, these two zones are collectively referred to as the target monitoring zone. To improve the adaptability of the alarm strategy for the alarm and buffer zones, based on the technical solution described in Example 1, we divide the alarm and buffer zones into multiple equidistant intervals, namely response segments. When a vehicle is predicted to be intruding, the alarm urgency and method are dynamically adjusted based on the response segment in which its current location is located. The specific solution is as follows:

[0167] Divide the target monitoring area into multiple response segments, assign warning weights to each response segment, and optimize the adaptability of the alarm strategy, including:

[0168] The target monitoring area is divided into response segments, each corresponding segment ;in, Respectively represent the target monitoring area position and the target monitoring area end position; ;

[0169] Assign a base weight to each segment ;

[0170] Get traffic density factor ;

[0171] Get visibility factor ;in, Indicates visibility;

[0172] Obtain historical accident rates ;in, represents the number of vehicles in each response segment, Indicates the density of labeled vehicles, Indicates the number of historical accidents in each response segment obtained from the database, It represents the total number of historical accidents in the target monitoring area obtained from the database;

[0173] The dynamic weight of each response segment ;in, They represent the traffic density sensitivity coefficient, the minimum time for a vehicle to reach the work area from its response section, the actual time for a vehicle to reach the work area from its response section, the visibility attenuation coefficient, and the accident intensification coefficient respectively;

[0174] Develop a segmented alarm strategy, including:

[0175] Divide alarm intensity levels according to dynamic weights ;

[0176] According to different alarm intensity levels, single-mode graded alarm is carried out, including:

[0177] if , then a single sound alarm will sound at the end point of the corresponding response segment;

[0178] if , then an intermittent sound alarm will be issued at the end point of the corresponding response segment;

[0179] if , a high-frequency sound alarm will be issued at the end point of the corresponding response segment.

[0180] For example, we set up five response segments within a distance of 0-2 km from the work area. An audible and visual alarm device was installed at the end of each response segment. Each response segment was assigned a different basic weight, such as 0.3, 0.5, 0.7, 0.9, and 1.0 from far to near. The basic weight reflects the spatial urgency of the alarm (that is, the closer to the work area, the higher the weight).

[0181] In the sunny and low-traffic road scenario, the lowest value of the intrusion probability with an alarm intensity of level 3 warning in the fifth section is greater than 0.7;

[0182] In foggy (low visibility) and highly congested scenarios, the dynamic weight of the fifth segment is 1.0×(1+0.3×2.5)×0.5 0.7 ≈1.8, the intrusion probability threshold for level 3 warning only needs to be reduced to no less than 0.39. That is to say, in foggy and highly congested scenarios (where intrusion is prone to occur), the increase in dynamic weight reduces the requirement for intrusion probability and improves the sensitivity of warning response.

[0183] Example 4:

[0184] During road construction, how to dynamically adjust the working status of each response section device to ensure monitoring results while saving equipment energy consumption. Based on Example 3, we propose the following technical solution:

[0185] Dynamically adjust the warning status based on vehicle intrusion probability, recovery behavior, and vehicle location, allocate computing resources and communication priority based on the warning status, and reduce warning delays, including:

[0186] Obtaining the status of each response segment, wherein the status of the response segment includes monitoring state, warning state, emergency state and mitigation state;

[0187] when When , the corresponding response segment is converted from monitoring state to warning state; increase computing resources, improve communication priority, and reduce communication transmission cycle; Indicates the distance between the vehicle and the working area; Indicates a high risk threshold;

[0188] when When the corresponding response segment is converted from the warning state to the mitigation state; the original communication priority and allocated computing resources are restored;

[0189] when When the state transitions, the corresponding response segment switches from the warning state to the emergency state; the monitoring equipment and alarm equipment are kept working, the vehicle data collected by the monitoring equipment in the preset time period before and after the state transition is saved and transmitted to the monitoring center; GPU acceleration is started, and an independent channel is allocated to transmit the risk avoidance instructions;

[0190] when When the corresponding response stage changes from the mitigation state to the monitoring state; the monitoring equipment is kept working and the alarm equipment is dormant. .

[0191] The present invention also provides a road site security monitoring system based on big data and artificial intelligence, which is used to execute the road site security monitoring method based on big data and artificial intelligence.

[0192] A computer-readable storage medium stores a computer program, which is executed by a processor to implement the above-mentioned road site security monitoring method based on big data and artificial intelligence.

[0193] In the embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the above-mentioned functions defined in the method of the present application are performed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wire segments, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a propagated data signal, either in baseband or as part of a carrier wave, embodying computer-readable program code. Such a propagated data signal may take various forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical fiber cable, RF, etc., or any suitable combination thereof.

[0194] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or portion of code that contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as combinations of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0195] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be subject to any changes or modifications.

Claims

1. A road security monitoring method based on big data and artificial intelligence, characterized in that: The method comprises: Collecting vehicle data entering the target monitoring area at a preset collection frequency, and using the vehicle data to monitor vehicle operation characteristics and driver status characteristics; Based on the vehicle operation characteristics and driver status characteristics, the pre-trained danger warning model predicts the vehicle intrusion probability and generates corresponding alarm information; including: Divide the target monitoring area into multiple response segments, assign warning weights to each response segment, and optimize the adaptability of the alarm strategy, including: The target monitoring area is divided into corresponding segments, each corresponding segment ;in, Respectively represent the target monitoring area position and the target monitoring area end position; ; Assign a base weight to each segment ; Get traffic density factor ; Get visibility factor ;in, Indicates visibility; Obtain historical accident rates ;in, represents the number of vehicles in each response segment, Indicates the density of labeled vehicles, Indicates the number of historical accidents in each response segment obtained from the database, It represents the total number of historical accidents in the target monitoring area obtained from the database; The dynamic weight of each response segment ;in, They represent the traffic density sensitivity coefficient, the minimum time for a vehicle to reach the work area from its response section, the actual time for a vehicle to reach the work area from its response section, the visibility attenuation coefficient, and the accident intensification coefficient respectively; Determine vehicle intrusion patterns, filter alarm information, and extract effective alarm information; After obtaining effective alarm information, a multimodal hierarchical alarm strategy is constructed to provide real-time feedback of target monitoring area image data and generate the optimal risk avoidance path.

2. The road site security monitoring method based on big data and artificial intelligence according to claim 1 is characterized in that: The vehicle data includes video data of vehicles running in the target monitoring area, infrared image data of drivers, vehicle speed data, vehicle identity data and passing time data; The monitoring of vehicle operation characteristics and driver status characteristics using the vehicle data includes: Align the video data and speed data of vehicles running in the target monitoring area on the time axis; Extract the vehicle's speed characteristics from the vehicle's speed data, including the vehicle's instantaneous speed and acceleration ; Extract vehicle direction change features, including heading angle deviation, from vehicle video data , lateral displacement rate and trajectory curvature ; in, , Indicates the direction of vehicle travel. Indicates lane direction; , , represents the vehicle displacement increment; Extracting driver status features, including distraction indicators, from driver infrared image data , head offset angle , blink frequency ;in, Indicates how long the eyes are looking at the road. Indicates the total fixation duration.

3. The road site security monitoring method based on big data and artificial intelligence according to claim 2 is characterized in that: The method of predicting the vehicle intrusion probability based on the vehicle operation characteristics and the driver status characteristics through the pre-trained danger warning model and generating corresponding alarm information includes: After normalizing the speed feature, direction change feature and driver state feature, the initial vehicle feature vector is constructed ;in They represent the normalized instantaneous velocity, acceleration, heading angle deviation, lateral displacement rate, trajectory curvature, Euclidean distance from the vehicle to the warning area boundary, and attention distraction index respectively; Construct a feature vector sequence within the acquisition time window ; Indicates the length of the acquisition time window; Indicates the end point of the collection period; Inputting the feature vector into a pre-trained risk prediction model, wherein the risk prediction model is a logistic regression model; Predicting vehicle intrusion probability through danger prediction model ,in, , represents the intercept, Indicates the Time point The weight of each feature; Indicates time No. Features Indicates the number of eigenvector elements; If the invasion probability When the value is greater than the preset alarm threshold, an alarm message is generated to remind on-site staff that there is a vehicle intrusion; The image data of the corresponding vehicle is obtained, and the image data of the vehicle is associated with the alarm information and sent to the cloud to remind the monitoring personnel that there is a vehicle intrusion.

4. The road site security monitoring method based on big data and artificial intelligence according to claim 3 is characterized in that: The method of predicting the vehicle intrusion probability based on the vehicle operation characteristics and the driver status characteristics through the pre-trained danger warning model and generating corresponding alarm information also includes: Obtain the motion characteristics and driver status characteristics of vehicles in adjacent lanes to the target monitoring area and predict vehicle lane change intrusion, including: Acquire running image data of vehicles in adjacent lanes, vehicle running speed and driver status image data; Align vehicle operation image data, vehicle operation speed, and driver status image data through hardware-triggered timestamp synchronization; Extract vehicle motion features in adjacent lanes, including lane line angles and the vehicle's relative position change rate ,in, Indicates the distance the vehicle moves along the lane line during the collection interval. Indicates the distance the vehicle moves in the direction perpendicular to the lane line during the collection interval; After normalizing the motion features of vehicles in adjacent lanes, a fusion vehicle feature vector is constructed. ;in, Represent the normalized lane line angle and vehicle relative position change rate respectively; Construct a fusion feature vector sequence within the acquisition time window ; The vehicle extended feature vector is input into the pre-trained enhanced intrusion risk prediction model, and the fused intrusion probability is output. ; Specifically include: An enhanced intrusion risk prediction model is constructed through a bidirectional LSTM network, wherein the enhanced intrusion risk prediction model includes a motion branch, a driver branch, and a decision fusion branch; Output motion feature vector through motion branch , output the driver feature vector through the driver branch ; Fusion invasion probability ;in, represents the output layer fusion matrix, represents the output layer bias; represents the motion feature fusion weight and driver feature fusion weight; If the fused intrusion probability is greater than the preset alarm threshold, an alarm message is output.

5. The road site security monitoring method based on big data and artificial intelligence according to claim 4 is characterized in that: The method of predicting the vehicle intrusion probability based on the vehicle operation characteristics and the driver status characteristics through the pre-trained danger warning model and generating corresponding alarm information also includes: Develop a segmented alarm strategy, including: Divide alarm intensity levels according to dynamic weights ; According to different alarm intensity levels, single-mode graded alarm is carried out, including: if , then a single sound alarm will sound at the end point of the corresponding response segment; if , then an intermittent sound alarm will be issued at the end point of the corresponding response segment; if , a high-frequency sound alarm will be issued at the end point of the corresponding response segment.

6. The road site security monitoring method based on big data and artificial intelligence according to claim 5 is characterized in that: The process of determining the vehicle intrusion mode, filtering the alarm information, and extracting the effective alarm information includes: Distinguish intrusion patterns in the lane and adjacent lanes within the target monitoring area, monitor vehicle exit behavior in real time, and optimize intrusion probability prediction, including: Perform lane classification and intrusion type determination, including: Encode lane space, lane coding ; if , then it is judged that the vehicle intrudes straight into the lane; if , then it is judged that the adjacent lane is intruded; if , it is judged that the vehicle is out of control and intrudes; Monitor the vehicle's operating status and determine whether the vehicle has changed out of the corresponding motion state, including: Set the recovery judgment indicator ;in, Indicates that the vehicle moves laterally in the original lane. Indicates that the vehicle is actively driving to avoid a maneuver; Calculate the corrected fusion invasion probability ,in, represents the probability attenuation coefficient; time interval; If the fusion intrusion probability is greater than the preset alarm threshold and When the vehicle changes its intrusion state, the alarm information is considered invalid and the alarm information output is stopped; If the fusion intrusion probability is greater than the preset alarm threshold and , it is determined that the vehicle has not changed the intrusion status, the alarm information is a valid alarm, and the alarm information is kept output.

7. The road site security monitoring method based on big data and artificial intelligence according to claim 6 is characterized in that: The multi-modal hierarchical alarm strategy is constructed to provide real-time feedback of target monitoring area image data, including: A multi-modal hierarchical alarm strategy is constructed based on the alarm intensity level and recovery judgment indicators, including: Obtain alarm intensity level, recovery judgment index, video stream of target monitoring area and video stream of working area; if and , sound and light alarm is performed in the last response segment of the target monitoring area; if and , sound and light alarms are issued in the working area, and alarm information and video streams of the target monitoring area are sent to the monitoring center, and intruding vehicles are marked in the video streams of the target monitoring area; if and If the time interval is greater than 1.5 seconds, an audible and visual alarm will be sounded in the working area, and a rescue request will be sent to the monitoring center; Generate the optimal risk avoidance path and feed it back to the monitoring center; Generating the optimal risk avoidance path includes: Obtain image information of the work area and the coordinate information of the staff from the GIS system. Combined with the predicted trajectory of the intruding vehicle, the optimal avoidance path is calculated, including: Predicting intruding vehicle trajectories: ;in, Represent the vehicle coordinates at the current moment, represents the prediction time interval, Indicates the acquisition frequency; Predict vehicle driving direction angle ; Obtain staff coordinate information and construct threat areas ; ,in, Indicates the radius of the safe area; Indicates any coordinate point within the threat area; Combine the vehicle's predicted direction angle to construct the optimal risk avoidance path: ,in, They represent the direction vector of the staff and the direction vector of the nearest safe haven respectively. Indicates the normalized value of the direction angle from the worker to the shelter; Represents the normalized value of the vehicle's predicted driving direction angle; Represent position weight and angle weight respectively; Indicates the coordinates of the nearest safe haven; get and Finally, an alternative escape path is generated through the Bezier curve, including: Direct safe-haven path ,in, represents the correction factor, ; Left avoidance path , where the lateral offset ; Right avoidance path ; Calculation acquisition 、 、 、 , generate alarm instructions and send them to the staff in the work area to guide them to take emergency precautions.

8. The road site security monitoring method based on big data and artificial intelligence according to claim 7 is characterized in that: It also includes dynamically adjusting the warning status based on vehicle intrusion probability, recovery behavior, and vehicle location, allocating computing resources and communication priority based on the warning status, and reducing warning delays, including: Obtaining the status of each response segment, wherein the status of the response segment includes monitoring state, warning state, emergency state and mitigation state; when When , the corresponding response segment switches from monitoring state to warning state; Indicates the distance between the vehicle and the working area; Indicates a high risk threshold; when When , the corresponding response segment switches from the warning state to the mitigation state; when When the corresponding response segment is converted from the warning state to the emergency state; when When the corresponding response stage changes from mitigation state to monitoring state, the low risk threshold .

9. The road site security monitoring system based on big data and artificial intelligence is characterized by: The system is used to execute the road site security monitoring method based on big data and artificial intelligence as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the road site security monitoring method based on big data and artificial intelligence as described in any one of claims 1 to 8.

Images