Road traffic event analysis method and system, and device and medium

Abstract

A road traffic event analysis method and system, and a device and a medium. The traffic event analysis method comprises: decoding and coding the current monitoring video, so as to obtain a target video picture (S102); calibrating the target video picture, so as to obtain a calibrated video picture (S103); performing preliminary analysis on the calibrated video picture, so as to obtain a preliminary analysis result (S104); determining state change data on the basis of the preliminary analysis result (S105); compiling statistics on a traffic flow, so as to determine peak time periods and plateau time periods of a road traffic site (S106); performing preliminary detection on frame scenes on the basis of pre-determined detection parameters corresponding to the peak time periods and the plateau time periods, so as to obtain a preliminary detection result (S107); excluding congestion events from the preliminary detection result (S108); and tracking targets, and determining motion state features, so as to obtain a final detection result and report same (S109). The scale of a traffic accident event triggered in the detection result and the statistical result are used for adjusting the detection parameters corresponding to the peak time periods and the plateau time periods. The traffic event analysis method can improve the detection accuracy of road traffic events.

Description

Road traffic incident analysis method, system, device and medium

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on December 29, 2023, with application number 202311866599.2 and invention name “A road traffic incident analysis method, system, equipment and medium”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present invention relates to the field of road traffic identification, and in particular to a road traffic incident analysis method, system, equipment and medium. Background Art

[0003] Urban expressways, highways, and tunnels, as convenient transportation options, face the challenge of balancing growing traffic demand with limited capacity. This leads to frequent incidents such as traffic accidents, vehicle breakdowns, and spilled debris, significantly reducing traffic efficiency and negatively impacting society. Therefore, employing event detection algorithms to promptly and accurately identify these traffic issues is crucial to ensuring the safe operation of urban expressways, highways, and tunnels, and ensuring smooth public travel.

[0004] The existing technology uses a self-tuning method for traffic event detection algorithms based on genetic algorithms. Using historical data and targeting the performance indicators of event detection, a genetic algorithm is used to optimize the threshold parameters of the detection algorithm and complete offline tuning of the threshold parameters. By continuously determining whether the event detection meets the trigger conditions within the cycle time, the relevant parameter factors are iterated again and again to optimize the algorithm threshold parameters. Existing event detection algorithms limit the detection function to detection within a certain cycle, that is, a certain time length is set and the detection cycle is continuously refreshed within this time window. There is no targeted optimization for event detection with longer or shorter time spans. At the same time, they rely heavily on historical data, but this historical data is often not so easy to see, so that in the absence of historical data, the detection accuracy may not be satisfactory when a new scene is changed.

[0005] Summary of the Invention

[0006] Based on this, embodiments of the present invention provide a road traffic incident analysis method, system, device, and medium to improve the detection accuracy of road traffic incidents.

[0007] To achieve the above objectives, the embodiments of the present invention provide the following solutions:

[0008] A road traffic incident analysis method, comprising:

[0009] Obtain current surveillance video of road traffic scenes;

[0010] The AI ​​video detection server is used to decode and re-encode the current surveillance video to obtain a target video image; the video format of the target video image is a video format suitable for AI video detection;

[0011] Calibrate the detection range of the target video image, the mapping relationship between the target video image and the latitude and longitude in the real world, and the event detection start time period to obtain a calibrated video image;

[0012] Perform a preliminary analysis on the calibrated video image to obtain a preliminary analysis result; the preliminary analysis result includes the detection image, vertex coordinates of the detection frame, the confidence of the detection frame, the license plate number, and the target tracking number;

[0013] Determine each independent target based on the preliminary analysis results, and continuously track each independent target to determine state change data of each independent target; the state change data includes speed and relative position;

[0014] Performing traffic flow statistics on the road traffic scene to obtain statistical results, and determining a peak time period and an off-peak time period of the road traffic scene based on the statistical results; the peak time period is a time period when the traffic flow is greater than or equal to a set flow value; and the off-peak time period is a time period when the traffic flow is less than the set flow value;

[0015] Based on the state change data, preliminary detection is performed on each frame scene in the target video according to predetermined detection parameters corresponding to a peak time period and detection parameters corresponding to an off-peak time period to obtain preliminary detection results; the preliminary detection results include whether there are relevant events at the road traffic scene; the relevant events include congestion events and traffic accident events;

[0016] Eliminating congestion events in the preliminary detection results according to the statistical results;

[0017] Track each target in the frame scene excluding congestion events and determine the motion state characteristics of each target;

[0018] The final detection result is determined based on the motion state characteristics, and the final detection result is reported to the event platform; the final detection result includes whether a traffic accident event is triggered; the scale of the triggered traffic accident event and the statistical results are used to adjust the detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period to perform the detection of the next monitoring video.

[0019] Optionally, calibrating the mapping relationship between the target video image and the latitude and longitude of the real world specifically includes:

[0020] For any frame of the target video screen, a set number of pixel points in the scene are selected as target points; the set number of pixel points are evenly distributed in the target video screen;

[0021] For any frame of the target video screen, the latitude and longitude coordinates corresponding to the target point in the scene are collected using a GPS acquisition tool, laser scanning, or a drone overlooking the target point, and the pixel coordinates of the target point and the latitude and longitude coordinates form a coordinate combination pair;

[0022] A mapping matrix for converting the camera coordinate system and the real world coordinate system is generated by combining the camera matrix, distortion coefficient, rotation matrix, and translation vector according to the coordinate combination of all frame scenes in the target video screen.

[0023] Optionally, a preliminary analysis is performed on the calibrated video image to obtain a preliminary analysis result, specifically including:

[0024] Each video stream in the calibrated video screen corresponds to a thread. The thread corresponding to each video stream gradually extracts frame data and analyzes each frame data extracted to obtain the frame result.

[0025] The frame results of all frame data of each video stream in the calibrated video picture are used as the preliminary analysis results;

[0026] Analyze any frame of data to obtain the frame results, including:

[0027] Sending the frame data to a decoder corresponding to the target thread using a target thread, where the target thread is the thread corresponding to the frame data;

[0028] After decoding the frame data, the decoder reduces the frame rate of the frame data according to the set detection requirements, and uses the data with the reduced frame rate as decoded image data;

[0029] The decoded image data is stored in a cache queue, a sending thread is initiated in the target thread, and the sending thread is used to send the data stored in the cache queue to the thread to be inferred in a first-in-first-out manner;

[0030] After the inference thread infers the received data, it puts the inference result into the cache queue; the inference result includes: the detection image, the vertex coordinates of the detection box and the confidence of the detection box;

[0031] License plate recognition and target tracking are performed according to the data stored in the cache queue to obtain the frame result.

[0032] Optionally, based on the state change data, preliminary detection is performed on each frame scene in the target video according to predetermined detection parameters corresponding to a peak time period and detection parameters corresponding to a non-peak time period to obtain preliminary detection results, specifically including:

[0033] For the frame scene corresponding to the peak time period, marking the objects in the parked state in the frame scene corresponding to the peak time period according to the state change data, and detecting the positions of the objects in the parked state;

[0034] Correcting the detection parameters corresponding to the peak time period according to the position of the target in the parking state to obtain a correction parameter;

[0035] Determine whether the number of objects in a stopped state and the speed of objects in a moving state meet the set conditions;

[0036] When set conditions are met, a stability coefficient is calculated based on the correction parameter, and whether a related event exists at the road traffic scene is determined based on the stability coefficient to obtain a preliminary detection result; the set conditions include: when the number of parked objects falls within the interval [2, 3], and when a moving object is located behind a parked object, the speed of the moving object is greater than 10 km / h; or when the number of parked objects falls within the interval [2, 3], and when a moving object is located in a lane surrounding the parked object, the speed of the moving object is greater than 20 km / h; the stability coefficient is used to indicate whether the related event has the possibility of continuing to occur within a set time period;

[0037] For the frame scene corresponding to the off-peak time period, predicting whether a path conflict will occur for a target in the frame scene corresponding to the off-peak time period based on the state change data;

[0038] If so, predict the time when the path conflict occurs, and obtain a high-definition surveillance video of the road traffic scene according to the time when the conflict occurs; the resolution and frame rate of the high-definition surveillance video are both greater than the resolution and frame rate of the current surveillance video;

[0039] Event detection is performed on the high-definition surveillance video using detection parameters corresponding to the off-peak time period to obtain preliminary detection results.

[0040] The present invention also provides a road traffic incident analysis system, comprising:

[0041] An acquisition module is used to obtain the current monitoring video of the road traffic scene;

[0042] A decoding and encoding module is used to decode and re-encode the current monitoring video using the AI ​​video detection server to obtain a target video image; the video format of the target video image is a video format suitable for AI video detection;

[0043] a calibration module, configured to calibrate the detection range of the target video image, the mapping relationship between the target video image and the longitude and latitude of the real world, and the event detection start time period, to obtain a calibrated video image;

[0044] A preliminary analysis module is used to perform preliminary analysis on the calibrated video image to obtain preliminary analysis results; the preliminary analysis results include the detection image, vertex coordinates of the detection frame, the confidence of the detection frame, the license plate number, and the target tracking number;

[0045] A state tracking module is used to determine each independent target based on the preliminary analysis results, and continuously track each independent target to determine the state change data of each independent target; the state change data includes speed and relative position;

[0046] a time period determination module, configured to perform traffic flow statistics on the road traffic scene, obtain statistical results, and determine a peak time period and an off-peak time period of the road traffic scene based on the statistical results; the peak time period is a time period when the traffic flow is greater than or equal to a set flow value; and the off-peak time period is a time period when the traffic flow is less than the set flow value;

[0047] a preliminary detection module configured to perform preliminary detection on each frame scene in the target video image based on the state change data and according to predetermined detection parameters corresponding to a peak time period and a detection parameter corresponding to an off-peak time period, to obtain preliminary detection results; the preliminary detection results include whether there are relevant events at the road traffic scene; the relevant events include congestion events and traffic accident events;

[0048] a congestion elimination module, configured to eliminate congestion events in the preliminary detection results according to the statistical results;

[0049] The target tracking module is used to track each target in the frame scene excluding the congestion event and determine the motion state characteristics of each target;

[0050] The final result determination module is used to determine the final detection result based on the motion state characteristics and report the final detection result to the event platform; the final detection result includes whether a traffic accident event is triggered; the scale of the triggered traffic accident event and the statistical results are used to adjust the detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period to perform the next monitoring video detection.

[0051] Optionally, the calibration module includes a detection range calibration submodule, a mapping relationship calibration submodule, and a time period calibration submodule; the detection range calibration submodule is used to calibrate the detection range of the target video image; the mapping relationship calibration submodule is used to calibrate the mapping relationship between the target video image and the longitude and latitude of the real world; the time period calibration submodule is used to calibrate the event detection start time period;

[0052] The mapping relationship calibration submodule includes:

[0053] A target point selection unit is used to select a set number of pixel points in any frame of the target video image as target points; the set number of pixel points are evenly distributed in the target video image;

[0054] A coordinate unit is used to collect the longitude and latitude coordinates corresponding to the target point in any frame of the target video screen using a GPS acquisition tool, laser scanning, or a drone overlooking the target point, and to form a coordinate combination pair with the pixel coordinates of the target point and the longitude and latitude coordinates;

[0055] A mapping matrix generation unit is used to generate a mapping matrix for converting the camera coordinate system and the real world coordinate system based on the camera matrix, distortion coefficient, rotation matrix and translation vector according to the coordinate combination of all frame scenes in the target video screen.

[0056] Optionally, the preliminary analysis module includes:

[0057] The frame processing submodule is used to correspond one thread to each video stream in the calibrated video screen. The corresponding thread of each video stream gradually extracts frame data and analyzes each extracted frame data to obtain the frame result;

[0058] A preliminary analysis result determination submodule is used to use the frame results of all frame data of each video stream in the calibrated video picture as the preliminary analysis result;

[0059] The frame processing submodule includes a single frame processing unit, which is used to analyze any frame data to obtain a frame result;

[0060] The single frame processing unit includes:

[0061] A sending subunit, configured to send the frame data to a decoder corresponding to the target thread using a target thread, where the target thread is the thread corresponding to the frame data;

[0062] a frame rate reduction subunit, configured to reduce the frame rate of the frame data according to a set detection requirement after the decoder decodes the frame data, and use the data with the reduced frame rate as decoded image data;

[0063] The thread sub-unit is configured to store the decoded image data into a cache queue, initiate a sending thread in the target thread, and use the sending thread to send the data stored in the cache queue to the thread to be inferred in a first-in-first-out manner;

[0064] An inference result subunit, configured to place the inference result into the cache queue after the inference thread infers the received data; the inference result includes: a detection image, vertex coordinates of a detection box, and a confidence level of the detection box;

[0065] The frame result subunit is used to perform license plate recognition and target tracking according to the data stored in the cache queue to obtain the frame result.

[0066] Optionally, the preliminary detection module includes:

[0067] a peak time mark detection unit, configured to mark objects in a parked state in the frame scene corresponding to the peak time period according to the state change data, and detect the positions of the objects in the parked state;

[0068] a correction unit, configured to correct the detection parameters corresponding to the peak time period according to the position of the object in the parked state, to obtain a correction parameter;

[0069] a judging unit, configured to judge whether the number of objects in a parked state and the speed of objects in a moving state meet set conditions;

[0070] a preliminary inspection unit, configured to calculate a stability coefficient based on the correction parameter when set conditions are met, and determine whether a related event exists at the road traffic scene based on the stability coefficient, thereby obtaining a preliminary inspection result; the set conditions comprising: the number of parked objects falling within the interval [2, 3], and when a moving object is located behind a parked object, the speed of the moving object is greater than 10 km / h; or the number of parked objects falling within the interval [2, 3], and when a moving object is located in a lane surrounding the parked object, the speed of the moving object is greater than 20 km / h; the stability coefficient is used to indicate whether the related event has the possibility of continuing to occur within a set time period;

[0071] a flat-peak period prediction unit, configured to predict, for a frame scene corresponding to the flat-peak period, based on the state change data, whether a path conflict will occur for a target in the frame scene corresponding to the flat-peak period;

[0072] A prediction result unit is configured to predict the time when the path conflict occurs, and obtain a high-definition monitoring video of the road traffic scene according to the time when the conflict occurs; the resolution and frame rate of the high-definition monitoring video are both greater than the resolution and frame rate of the current monitoring video;

[0073] The off-peak period event detection unit is used to perform event detection on the high-definition surveillance video using detection parameters corresponding to the off-peak period to obtain preliminary detection results.

[0074] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned road traffic incident analysis method.

[0075] The present invention also provides a computer-readable storage medium storing a computer program, which implements the above-mentioned road traffic incident analysis method when executed by a processor.

[0076] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0077] The embodiments of the present invention propose a road traffic event analysis method, system, equipment and medium, which perform preliminary detection on each frame scene in the target video image according to the detection parameters corresponding to the predetermined peak time period and off-peak time period, exclude congestion events in the preliminary detection results according to the traffic flow statistics results, and track each target in the frame scene excluding the congestion events, determine the motion state characteristics of each target, and thus determine the final detection result and report it to the event platform, and adjust the scale of the traffic accident event triggered in the final detection result and the detection parameters corresponding to the peak time period and the off-peak time period of the traffic flow statistics results. Compared with the existing event detection algorithm, there is no need to limit the detection function to detection within a certain period, nor to rely on a large amount of historical data, thereby improving the detection accuracy of road traffic events. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0079] FIG1 is a flowchart of a road traffic incident analysis method according to an embodiment of the present invention;

[0080] FIG2 is a second flowchart of the road traffic incident analysis method provided by an embodiment of the present invention;

[0081] FIG3 is a structural diagram of a road traffic incident analysis system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings in the embodiments of the present invention.

[0083] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0084] Example 1

[0085] Referring to FIG1 , the road traffic incident analysis method of this embodiment includes:

[0086] Step 101: Acquire the current surveillance video of the road traffic scene.

[0087] Step 102: Use the AI ​​video detection server to decode and re-encode the current surveillance video to obtain the target video image.

[0088] Among them, the video format of the target video picture is a video format suitable for AI video detection.

[0089] Step 103: Calibrate the detection range of the target video image, the mapping relationship between the target video image and the longitude and latitude of the real world, and the event detection start time period to obtain a calibrated video image.

[0090] Step 104: Perform a preliminary analysis on the calibrated video image to obtain a preliminary analysis result.

[0091] The preliminary analysis results include the detection image, vertex coordinates of the detection frame, the confidence of the detection frame, the license plate number, and the target tracking number.

[0092] Step 105: Determine each independent target based on the preliminary analysis results, and continuously track each independent target to determine the state change data of each independent target; the state change data includes speed and relative position.

[0093] Step 106: performing traffic flow statistics on the road traffic scene to obtain statistical results, and determining the peak time period and off-peak time period of the road traffic scene based on the statistical results.

[0094] The peak time period is a time period when the traffic flow is greater than or equal to a set flow value; the off-peak time period is a time period when the traffic flow is less than a set flow value.

[0095] Step 107: Based on the state change data, preliminary detection is performed on each frame scene in the target video according to predetermined detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period to obtain preliminary detection results.

[0096] The preliminary detection result includes whether there are relevant events at the road traffic scene; the relevant events include congestion events and traffic accident events.

[0097] Step 108: excluding congestion events in the preliminary detection results according to the statistical results, tracking each target in the frame scene excluding the congestion event, and determining the motion state characteristics of each target.

[0098] Step 109: Determine the final detection result based on the motion state characteristics, and report the final detection result to the event platform.

[0099] Among them, the final detection result includes whether a traffic accident event is triggered; the scale of the triggered traffic accident event and the statistical results are used to adjust the detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period to perform the next monitoring video detection.

[0100] In one example, step 103 specifically includes:

[0101] For any frame of the target video screen, a set number of pixel points in the scene are selected as target points; the set number of pixel points are evenly distributed in the target video screen.

[0102] For any frame of the target video screen, the longitude and latitude coordinates corresponding to the target point in the scene are collected using GPS collection tools, laser scanning or drone viewing according to the target point, and the pixel coordinates of the target point and the longitude and latitude coordinates form a coordinate combination pair.

[0103] A mapping matrix for converting the camera coordinate system and the real world coordinate system is generated by combining the camera matrix, distortion coefficient, rotation matrix, and translation vector according to the coordinate combination of all frame scenes in the target video screen.

[0104] In one example, step 104 specifically includes:

[0105] Each video stream in the calibrated video screen corresponds to a thread. The thread corresponding to each video stream gradually extracts frame data and analyzes each extracted frame data to obtain a frame result.

[0106] The frame results of all frame data of each video stream in the calibrated video picture are used as the preliminary analysis results.

[0107] Among them, any frame of data is analyzed to obtain the frame result, which specifically includes:

[0108] The frame data is sent to a decoder corresponding to the target thread using a target thread, where the target thread is a thread corresponding to the frame data.

[0109] After decoding the frame data, the decoder reduces the frame rate of the frame data according to the set detection requirements, and uses the data with the reduced frame rate as decoded image data.

[0110] The decoded image data is stored in a cache queue, a sending thread is initiated in the target thread, and the sending thread is used to send the data stored in the cache queue to the thread to be inferred in a first-in-first-out manner.

[0111] After the thread to be inferred has inferred a result on the received data, it puts the inference result into the cache queue; the inference result includes: a detection image, vertex coordinates of a detection box, and a confidence level of the detection box.

[0112] License plate recognition and target tracking are performed according to the data stored in the cache queue to obtain the frame result.

[0113] In one example, step 107 specifically includes:

[0114] (1) For the frame scene corresponding to the peak time period, the objects in the parking state in the frame scene corresponding to the peak time period are marked according to the state change data, and the positions of the objects in the parking state are detected.

[0115] The detection parameters corresponding to the peak time period are corrected according to the position of the target in the parking state to obtain the corrected parameters.

[0116] It is determined whether the number of objects in the parked state and the speed of objects in the moving state meet the set conditions.

[0117] When the set conditions are met, the stability coefficient is calculated according to the correction parameters, and whether there are related events at the road traffic scene is determined according to the stability coefficient to obtain a preliminary detection result; the set conditions include: when the number of targets in a parked state falls within the interval [2, 3] and the target in motion is located behind the target in a parked state, the speed of the target in motion is greater than 10 km / h; or, when the number of targets in a parked state falls within the interval [2, 3] and the target in motion is located in the surrounding lane of the target in a parked state, the speed of the target in motion is greater than 20 km / h; the stability coefficient is used to characterize whether the related events have the possibility of continuing to occur within a set time period.

[0118] (2) For the frame scene corresponding to the off-peak time period, based on the state change data, it is predicted whether a path conflict will occur to the target in the frame scene corresponding to the off-peak time period.

[0119] If so, the time when the path conflict occurs is predicted, and a high-definition monitoring video of the road traffic scene is obtained according to the time when the conflict occurs; the resolution and frame rate of the high-definition monitoring video are both greater than the resolution and frame rate of the current monitoring video.

[0120] Event detection is performed on the high-definition surveillance video using detection parameters corresponding to the off-peak time period to obtain preliminary detection results.

[0121] A more specific implementation process of the above-mentioned road traffic incident analysis method in practical application is given below, and the method is further explained in detail.

[0122] The road traffic incident analysis method of this specific example mainly includes data access, detection scene calibration, target recognition, trajectory analysis, parameter adaptive adjustment, and event recognition, which specifically includes the following steps:

[0123] Step 1: Connect the surveillance video of the traffic scene to the AI ​​video detection server in real time, decode the video, and re-encode it into a video format more suitable for AI detection.

[0124] Step 2: Perform simple calibration on the incoming video, including the detection range of the video screen, the mapping relationship between the video screen and the longitude and latitude of the real world, the time period for event detection, etc.

[0125] Step 3: Analyze the video image to identify basic information such as the type, location, and motion status of each target in the image.

[0126] Step 4: Analyze the data collected in Step 3 as a whole, linking them to individual targets and continuously tracking their status changes, including speed and position relative to the image. Traffic flow statistics are also collected for the detection scene, using statistical methods to determine the peak and off-peak time periods, as well as lane occupancy and time occupancy.

[0127] Step 5: Analyze the scene based on the data from step 4. If there are a large number of stationary objects in the scene, and if there are still a large number of stationary objects at the end of the image, a preliminary analysis is conducted to determine whether there is congestion or traffic accidents, and a preliminary report is made to the test platform.

[0128] Step 6: If the test platform corrects the accuracy of the initially reported event, the event is analyzed based on the actual situation and the results are fed back into the algorithm. After multiple feedback loops, the distance parameter D is automatically calculated to determine how to define distant and near objects in the current scene. If the test platform does not correct the accuracy of the initially reported event, this step is skipped and the existing historical parameters are used.

[0129] Step 7: Comprehensively analyze the traffic conditions in step 4 and the event information reported in step 5. After excluding the congestion event in the current scene, track each target in the scene throughout the entire process, comprehensively analyze the characteristics of each target's motion state, and dynamically adjust different detection parameters in the same scene throughout the day based on the traffic status data in step 4.

[0130] Step 8: After comprehensive analysis of scene features in step 7, if an event is detected, a special algorithm path for the event is entered, different parameter values ​​are adjusted according to the scale of the event, and the detection results are output.

[0131] Step 9. After confirmation, return the test results and report the event to the platform.

[0132] The main steps involved in the above steps are introduced in detail.

[0133] The process of preliminary calibration of the video image:

[0134] 1. Take a certain frame of the video stream, select points with significant features in the scene and evenly distribute them in the video screen, ensuring that there are coordinate (pixel) points of the same proportion at near and far distances, and the number is N. p ≥9 (pieces).

[0135] 2. Based on the point selection in step 1, use GPS acquisition tools, laser scanning or drone to collect some latitude and longitude coordinates N in the scene that correspond to step 1 one by one g, and generate pixel coordinates, latitude and longitude coordinates combination pair N (p|g) ={N p |N g}.

[0136] 3. The mapping matrix for mutual conversion between the two coordinate systems is generated by combining the conversion formula with the camera matrix, distortion coefficient, rotation matrix, translation vector, etc., so as to realize the function of obtaining the longitude and latitude coordinates from the pixel coordinates of the camera image.

[0137] The process of preliminary analysis of the picture:

[0138] 1. Open the target channel video or picture to be detected, and draw the target detection area at the corresponding position (to remove targets in the non-detection area and reduce false detection).

[0139] 2. When the system starts, open the video stream of the target channel to be detected and pull the stream in real time. If there are multiple video streams, multiple threads will open different channels respectively.

[0140] 3. In each thread that opens the stream, frame data is gradually obtained; each frame data is sent to the decoder corresponding to this thread; after decoding, the decoder reduces the frame rate according to the detection requirements. If the original frame rate is 25, the frame rate is reduced by half.

[0141] 4. The decoded image data is stored in a cache queue. A separate sending thread is started within this thread, and the data in the cache queue is sent to the inference thread in a first-in, first-out manner. (The purpose of starting a separate sending thread to send data is to avoid delays and jitter caused by sending data in the stream pulling thread, which affects the quality of the stream pulling data.)

[0142] 5. After the result is inferred, it is placed in a cache queue. The cache queue result contains the detection image, vertex coordinates of the detection box, and confidence level. The data in the cache queue is used as input for license plate recognition and multi-target tracking.

[0143] 6. Take a frame of data from the cache queue, perform license plate recognition, and update the result to a frame of data in the cache queue.

[0144] 7. Take out 10 frames of data from the cache queue (if less than 10 frames, take out all the data) as the input of multi-target tracking. Multi-target tracking is multi-threaded.

[0145] 8. Obtain the frame results of multi-target tracking, which include the detection image, vertex coordinates of the detection frame, confidence level, license plate number, target tracking number, etc. Next, perform construction detection and judgment processing on this data.

[0146] Target tracking and detection process:

[0147] 1. Save the results of each frame of video detection as an object list, including the ID, classification, confidence level, and pixel coordinates of the rectangular border of each detected object.

[0148] 2. Associate the objects in different frames and obtain the trajectories Γ = {T1, ..., T n}, each T i They all contain the target ID, category, frame number of the first appearance of the corresponding track, and whether the target is in the scene field of view.

[0149] 3. For a frame at a certain time t, calculate the IOU of the bounding box of all active trajectories at time tk (k≤30) and the bounding box at time t. If it is greater than or equal to the specified lower threshold, the current detected object is considered to match the corresponding object in trajectory Γ. Once a match is found, the object will be jumped out in the next k value cycle.

[0150] 4. After completing k cycles, if an object does not match any trajectory, it is considered a new target; if an object exceeds k max If the frame does not appear, it is determined that the object is not in the detection scene.

[0151] Multi-vehicle accident detection process:

[0152] 1. First, determine the peak hours and off-peak hours.

[0153] 2. During peak hours, all vehicles that have stayed in the image for more than T′ are marked as parked. The position of the parked vehicle is detected, and T is corrected by calculating its independent parameter Δt according to the formula based on the position, i.e., T = T′ + Δt.

[0154] 3. The number of parked vehicles in the current scene. When the number is within the range of [2, 3], analyze the speeds of other nearby vehicles. When a moving vehicle is behind a parked vehicle, detect whether its speed is greater than 10 km / h. When a moving vehicle is in the lane surrounding a parked vehicle, detect whether its speed is greater than 20 km / h.

[0155] 4. When the relevant conditions of step 3 are met, continue to observe the subsequent development trend of the event. Through parameter fitting, analyze whether the current event has the possibility of continuing to occur for a long time, that is, the stability coefficient δ, and decide whether to report the event.

[0156] 5. During off-peak hours, first predict the possibility of a collision between vehicles in the scene. Let t1 and t2 be the times when vehicles 1 and 2 arrive at the collision location, v1 and a1 be the current speed and acceleration of vehicle 1, v2 and a2 be the current speed and acceleration of vehicle 2, and R be the radius determined by the vehicle length L and width W:

[0157] 6. When the value of R is less than the normal distance between targets, an accident is considered likely. Further deduction shows that when t1 = t2, the collision time can be calculated. Based on the predicted time, the current channel is locked in advance, and more computing power is allocated to it. Using a higher-resolution, higher-frame-rate video stream, the entire detection scenario is analyzed for the accident, ensuring accuracy and reducing false alarms.

[0158] The flowchart of the above specific process is shown in Figure 2.

[0159] For urban roads, expressways, highways, and tunnels, the following challenges need to be addressed: 1. Addressing the difficulty of radar distinguishing between various types of vehicles; 2. Extracting the trajectory of the corresponding target from known data, further analyzing the likelihood of an accident, and promptly reporting it to the platform. Existing methods for vehicle trajectory and accident detection are rarely used and have low accuracy. The method described in the above embodiment addresses the challenges of computer vision accident detection in video streams, facilitating its application in a wide range of scenarios.

[0160] The aforementioned road traffic incident analysis method is a global, dynamic, and adaptive parameter event analysis algorithm that can detect and report traffic incidents such as multi-vehicle accidents and vehicle breakdowns. This method reduces false alarm rates while minimizing missed detections, improving data quality and reducing the workload for subsequent data cleaning, analysis, and processing. This method is also illustrated using multi-vehicle accidents as an example. Based on this algorithmic approach, it can also reduce false alarm rates and missed detection rates for incidents such as single-vehicle breakdowns, spilled objects, vehicles driving the wrong way, reversing, and entering restricted areas.

[0161] Example 2

[0162] In order to execute the method corresponding to the above-mentioned embodiment 1 and achieve the corresponding functions and technical effects, a road traffic incident analysis system is provided below.

[0163] Referring to FIG3 , the system includes:

[0164] The acquisition module 201 is used to acquire the current monitoring video of the road traffic scene.

[0165] The decoding and encoding module 202 is used to use the AI ​​video detection server to decode and re-encode the current monitoring video to obtain a target video picture; the video format of the target video picture is a video format suitable for AI video detection.

[0166] The calibration module 203 is used to calibrate the detection range of the target video image, the mapping relationship between the target video image and the longitude and latitude of the real world, and the event detection start time period to obtain a calibrated video image.

[0167] The preliminary analysis module 204 is used to perform preliminary analysis on the calibrated video image to obtain preliminary analysis results; the preliminary analysis results include the detection image, vertex coordinates of the detection frame, the confidence of the detection frame, the license plate number and the target tracking number.

[0168] The state tracking module 205 is used to determine each independent target based on the preliminary analysis results, and continuously track each independent target to determine the state change data of each independent target; the state change data includes speed and relative position.

[0169] The time period determination module 206 is used to perform traffic flow statistics on the road traffic scene, obtain statistical results, and determine the peak time period and off-peak time period of the road traffic scene based on the statistical results; the peak time period is a time period when the traffic flow is greater than or equal to a set flow value; the off-peak time period is a time period when the traffic flow is less than the set flow value.

[0170] The preliminary detection module 207 is used to perform preliminary detection on each frame scene in the target video image based on the state change data and the detection parameters corresponding to the predetermined peak time period and the detection parameters corresponding to the off-peak time period to obtain preliminary detection results; the preliminary detection results include whether there are relevant events at the road traffic scene; the relevant events include congestion events and traffic accident events.

[0171] The congestion elimination module 208 is configured to eliminate congestion events in the preliminary detection results according to the statistical results.

[0172] The target tracking module 209 is used to track each target in the frame scene excluding the congestion event and determine the motion state characteristics of each target.

[0173] The final result determination module 210 is used to determine the final detection result based on the motion state characteristics and report the final detection result to the event platform; the final detection result includes whether a traffic accident event is triggered; the scale of the triggered traffic accident event and the statistical results are used to adjust the detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period to perform the next monitoring video detection.

[0174] In one example, the calibration module 203 includes a detection range calibration submodule, a mapping relationship calibration submodule and a time period calibration submodule; the detection range calibration submodule is used to calibrate the detection range of the target video image; the mapping relationship calibration submodule is used to calibrate the mapping relationship between the target video image and the longitude and latitude of the real world; the time period calibration submodule is used to calibrate the time period for event detection to start.

[0175] The mapping relationship calibration submodule includes:

[0176] The target point selection unit is used to select a set number of pixel points in any frame scene in the target video picture as target points; the set number of pixel points are evenly distributed in the target video picture.

[0177] A coordinate unit is used to collect the longitude and latitude coordinates corresponding to the target point in any frame of the target video screen by using a GPS collection tool, laser scanning or a drone overlooking the target point, and to form a coordinate combination pair with the pixel coordinates of the target point and the longitude and latitude coordinates.

[0178] A mapping matrix generation unit is used to generate a mapping matrix for converting the camera coordinate system and the real world coordinate system based on the camera matrix, distortion coefficient, rotation matrix and translation vector according to the coordinate combination of all frame scenes in the target video screen.

[0179] In one example, the preliminary analysis module 204 includes:

[0180] The frame processing submodule is used to correspond one thread to each video stream in the calibrated video screen. The corresponding thread of each video stream gradually extracts frame data and analyzes each extracted frame data to obtain a frame result.

[0181] The preliminary analysis result determination submodule is used to use the frame results of all frame data of each video stream in the calibrated video picture as the preliminary analysis result.

[0182] The frame processing submodule includes a single frame processing unit, and the single frame processing unit is used to analyze any frame data to obtain a frame result.

[0183] The single frame processing unit includes:

[0184] The sending subunit is used to send the frame data to a decoder corresponding to the target thread using a target thread, where the target thread is the thread corresponding to the frame data.

[0185] The frame rate reduction subunit is used to reduce the frame rate of the frame data according to the set detection requirements after the decoder decodes the frame data, and use the data with the reduced frame rate as decoded image data.

[0186] The thread subunit is used to store the decoded image data into a cache queue, initiate a sending thread in the target thread, and use the sending thread to send the data stored in the cache queue to the thread to be inferred in a first-in-first-out manner.

[0187] The inference result subunit is used to put the inference result into the cache queue after the inference thread infers the result of the received data; the inference result includes: the detection image, the vertex coordinates of the detection box and the confidence of the detection box.

[0188] The frame result subunit is used to perform license plate recognition and target tracking according to the data stored in the cache queue to obtain the frame result.

[0189] In one example, the preliminary detection module 207 includes:

[0190] The peak time mark detection unit is used to mark the objects in the parking state in the frame scenes corresponding to the peak time period according to the state change data, and detect the positions of the objects in the parking state.

[0191] The correction unit is used to correct the detection parameters corresponding to the peak time period according to the position of the target in the parking state to obtain the correction parameters.

[0192] The judging unit is used to judge whether the number of objects in the parked state and the speed of objects in the moving state meet set conditions.

[0193] The initial inspection unit is used to calculate the stability coefficient according to the correction parameter when the set conditions are met, and determine whether there is a related event at the road traffic scene according to the stability coefficient to obtain a preliminary detection result; the set conditions include: when the number of targets in a parked state falls within the interval [2, 3], and the target in motion is located behind the target in a parked state, the speed of the target in motion is greater than 10 km / h; or, when the number of targets in a parked state falls within the interval [2, 3], and the target in motion is located in the surrounding lane of the target in a parked state, the speed of the target in motion is greater than 20 km / h; the stability coefficient is used to characterize whether the related event has the possibility of continuing to occur within a set time period.

[0194] The off-peak period prediction unit is used to predict, for a frame scene corresponding to the off-peak period, whether a target in the frame scene corresponding to the off-peak period will have a path conflict based on the state change data.

[0195] The prediction result unit is used to predict the time when the path conflict occurs and obtain a high-definition monitoring video of the road traffic scene according to the time when the conflict occurs; the resolution and frame rate of the high-definition monitoring video are both greater than the resolution and frame rate of the current monitoring video.

[0196] The off-peak period event detection unit is used to perform event detection on the high-definition surveillance video using detection parameters corresponding to the off-peak period to obtain preliminary detection results.

[0197] Example 3

[0198] This embodiment provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the road traffic incident analysis method of the first embodiment.

[0199] Optionally, the above-mentioned electronic device may be a server.

[0200] In addition, an embodiment of the present invention further provides a computer-readable storage medium storing a computer program, which implements the road traffic incident analysis method of embodiment 1 when executed by a processor.

[0201] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0202] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for analyzing road traffic events, characterized in that, Including: Obtain the current surveillance video of the road traffic scene; Use an AI video detection server to decode and re-encode the current surveillance video to obtain a target video frame; The video format of the target video frame is a video format suitable for AI video detection; Calibrate the detection range of the target video frame, the mapping relationship between the target video frame and the longitude and latitude in the real world, and the event detection start time period to obtain a calibrated video frame; Perform a preliminary analysis on the calibrated video frame to obtain a preliminary analysis result; the preliminary analysis result includes detection pictures, the vertex coordinates of the detection frame, the confidence level of the detection frame, the license plate number, and the target tracking number; Determine each independent target according to the preliminary analysis result, and continuously track each independent target to determine the status change data of each independent target; The status change data includes speed and relative position; Count the traffic flow of the road traffic scene to obtain a statistical result, and determine the peak time period and off-peak time period of the road traffic scene according to the statistical result; the peak time period is the time period when the traffic flow is greater than or equal to the set flow value; the off-peak time period is the time period when the traffic flow is less than the set flow value; Based on the status change data, perform a preliminary detection on each frame scene in the target video frame according to the detection parameters corresponding to the pre-determined peak time period and the detection parameters corresponding to the off-peak time period to obtain a preliminary detection result; the preliminary detection result includes whether there is a relevant event in the road traffic scene; the relevant events include congestion events and traffic accident events; Exclude congestion events from the preliminary detection result according to the statistical result; Track each target in the frame scene after excluding congestion events to determine the motion state characteristics of each target; Determine the final detection result according to the motion state characteristics and report the final detection result to the event platform; the final detection result includes whether a traffic accident event is triggered; Trigger The scale of the traffic accident event and the statistical result are used to adjust the detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period for the detection of the next surveillance video.

2. The road traffic event analysis method according to claim 1, wherein, Calibrate the mapping relationship between the target video frame and the longitude and latitude in the real world, specifically including: For any frame scene in the target video frame, select a set number of pixel points in the scene as target points; the set number of pixel points are evenly distributed in the target video frame; For any frame scene in the target video frame, use a GPS acquisition tool, laser scanning or drone overhead view method to collect the longitude and latitude coordinates corresponding to the target points in the scene according to the target points, and form a coordinate combination pair of the pixel coordinates of the target points and the longitude and latitude coordinates; Generate a mapping matrix for converting the camera coordinate system and the real world coordinate system according to the coordinate combination pairs of all frame scenes in the target video frame, the camera matrix, the distortion coefficient, the rotation matrix, and the translation vector.

3. The road traffic event analysis method according to claim 1, wherein Perform a preliminary analysis on the calibrated video frame to obtain a preliminary analysis result, specifically including: Each video stream in the calibrated video frame corresponds to a thread. For each video stream's corresponding thread, frame data is gradually extracted, and each extracted frame of data is analyzed to obtain a frame result; The frame results of all frame data of each video stream in the calibrated video frame are used as the preliminary analysis results; Among them, analyzing any frame of data to obtain a frame result specifically includes: Sending this frame of data to the decoder corresponding to the target thread by the target thread, where the target thread is the thread corresponding to this frame of data; After decoding this frame of data by the decoder, reducing the frame rate of this frame of data according to the set detection requirements, and using the data with the reduced frame rate as the decoded image data; Storing the decoded image data in a cache queue and initiating A sending thread, and sending the data stored in the cache queue to the thread to be inferred in a first-in, first-out manner by the sending thread; After the thread to be inferred infers a result from the received data, putting the inference result into the cache queue; the inference result includes: the detected picture, the vertex coordinates of the detection box, and the confidence of the detection box; Performing license plate recognition and target tracking based on the data stored in the cache queue to obtain the frame result.

4. The road traffic event analysis method according to claim 1, wherein Based on the state change data, performing preliminary detection on each frame scene in the target video frame according to the detection parameters corresponding to the predetermined peak time period and the detection parameters corresponding to the off-peak time period to obtain preliminary detection results, specifically including: For the frame scene corresponding to the peak time period, according to the state change data, marking the targets in the frame scene corresponding to the peak time period that are in a parking state, and detecting the positions where the targets in the parking state are located; Correcting the detection parameters corresponding to the peak time period according to the positions where the targets in the parking state are located to obtain correction parameters; Judging whether the number of targets in the parking state and the speed of the targets in the moving state meet the set conditions; When the set conditions are met, calculating a stability coefficient according to the correction parameters, and determining whether there are relevant events at the road traffic scene according to the stability coefficient to obtain preliminary detection results; the set conditions include: when the number of targets in the parking state belongs to the interval [2, 3] and the targets in the moving state are behind the targets in the parking state, the speed of the targets in the moving state is greater than 10 km / h; or, when the number of targets in the parking state belongs to the interval [2, 3] and the targets in the moving state are in the surrounding lanes of the targets in the parking state, the speed of the targets in the moving state is greater than 20 km / h; the stability coefficient is used to characterize the possibility that the relevant events will continue to occur within a set time period; For the frame scene corresponding to the off-peak time period, predicting whether there will be path conflicts among the targets in the frame scene corresponding to the off-peak time period according to the state change data; If so, predict the time when a path conflict occurs, and obtain the high-definition surveillance video of the road traffic scene according to the time of the conflict; the resolution and frame rate of the high-definition surveillance video are both greater than those of the current surveillance video; Perform event detection on the high-definition surveillance video using the detection parameters corresponding to the off-peak time period to obtain a preliminary detection result.

5. A road traffic event analysis system, characterized in that Including: An acquisition module, configured to acquire the current surveillance video of the road traffic scene; A decoding and encoding module, configured to decode and re-encode the current surveillance video using an AI video detection server to obtain a target video frame; The video format of the target video frame is a video format suitable for AI video detection; A calibration module, configured to calibrate the detection range of the target video frame, the mapping relationship between the target video frame and the longitude and latitude of the real world, and the event detection start time period to obtain a calibrated video frame; A preliminary analysis module, configured to perform a preliminary analysis on the calibrated video frame to obtain a preliminary analysis result; the preliminary analysis result includes detection pictures, the vertex coordinates of the detection frame, the confidence level of the detection frame, the license plate number, and the target tracking number; A status tracking module, configured to determine each independent target according to the preliminary analysis result, and continuously track each independent target to determine the status change data of each independent target; The status change data includes speed and relative position; A time period determination module, configured to perform a traffic flow statistics on the road traffic scene to obtain a statistical result, and determine the peak time period and off-peak time period of the road traffic scene according to the statistical result; the peak time period is the time period when the traffic flow is greater than or equal to the set flow value; The off-peak time period is the time period when the traffic flow is less than the set flow value; A preliminary detection module, configured to perform a preliminary detection on each frame scene in the target video frame based on the status change data according to the detection parameters corresponding to the pre-determined peak time period and the detection parameters corresponding to the off-peak time period to obtain a preliminary detection result; the preliminary detection result includes whether there is a relevant event in the road traffic scene; the relevant events include congestion events and traffic accident events; A congestion exclusion module, configured to exclude the congestion events in the preliminary detection result according to the statistical result ; A target tracking module, configured to track each target in the frame scene excluding the congestion event to determine the motion state characteristics of each target; A final result determination module, configured to determine the final detection result according to the motion state characteristics and report the final detection result to the event platform; the final detection result includes whether a traffic accident event is triggered; The scale of the triggered traffic accident event and the statistical result are used to adjust the detection parameters corresponding to the peak time period and the detection parameters corresponding to the off-peak time period for the detection of the next surveillance video.

6. The road traffic event analysis system according to claim 5, wherein The calibration module includes a detection range calibration sub-module, a mapping relationship calibration sub-module, and a time period calibration sub-module; the detection range calibration sub-module is used to calibrate the detection range of the target video frame; the mapping relationship calibration sub-module is used to calibrate the mapping relationship between the target video frame and the longitude and latitude in the real world; the time period calibration sub-module is used to calibrate the start time period of event detection; The mapping relationship calibration sub-module includes: A target point selection unit, which is used to select a set number of pixel points in any frame scene of the target video frame as target points; the set number of pixel points are evenly distributed in the target video frame; A coordinate unit, which is used to collect the longitude and latitude coordinates corresponding to the target points in the scene of any frame of the target video frame by using a GPS acquisition tool, laser scanning, or drone overlooking according to the target points, and form a coordinate combination pair of the pixel coordinates and the longitude and latitude coordinates of the target points; A mapping matrix generation unit, which is used to generate a mapping matrix for converting the camera coordinate system and the real world coordinate system according to the coordinate combination pairs of all frame scenes in the target video frame, the camera matrix, the distortion coefficient, the rotation matrix, and the translation vector.

7. The road traffic event analysis system according to claim 5, characterized in that, The preliminary analysis module includes: A frame processing sub-module, where each video stream in the calibrated video frame corresponds to a thread. For each thread corresponding to a video stream, frame data is gradually extracted, and for each frame of data is analyzed to obtain a frame result; A preliminary analysis result determination sub-module, which is used to use the frame results of all frame data of each video stream in the calibrated video frame as the preliminary analysis result; The frame processing sub-module includes a single-frame processing unit, and the single-frame processing unit is used to analyze any frame of data to obtain a frame result; The single-frame processing unit includes: A sending sub-unit, which is used to send the frame data to the decoder corresponding to the target thread by using the target thread, and the target thread is the thread corresponding to the frame data; A frame rate reduction sub-unit, which is used to reduce the frame rate of the frame data according to the set detection requirements after the decoder decodes the frame data, and use the data after the frame rate reduction as the decoded image data; A thread sub-unit, which is used to store the decoded image data in a cache queue, initiate a sending thread in the target thread, and send the data stored in the cache queue to the thread to be inferred in a first-in-first-out manner by using the sending thread; An inference result sub-unit, which is used to put the inference result into the cache queue after the thread to be inferred infers the result from the received data; the inference result includes: the detected picture, the vertex coordinates of the detection box, and the confidence level of the detection box; A frame result sub-unit, which is used to perform license plate recognition and target tracking according to the data stored in the cache queue to obtain the frame result.

8. The road traffic event analysis system according to claim 5, characterized in that The preliminary detection module includes: The peak-hour marking detection unit is configured to, for the frame scene corresponding to the peak time period, mark the targets in the parking state in the frame scene corresponding to the peak time period according to the state change data, and detect the positions of the targets in the parking state; The correction unit is configured to correct the detection parameters corresponding to the peak time period according to the positions of the targets in the parking state to obtain correction parameters; The judgment unit is configured to judge whether the number of targets in the parking state and the speeds of the targets in the moving state meet the set conditions; The preliminary detection unit is configured to, when the set conditions are met, calculate a stability coefficient according to the correction parameters, and determine whether there is a relevant event at the road traffic scene according to the stability coefficient to obtain a preliminary detection result; the set conditions include: when the number of targets in the parking state belongs to the interval [2, 3] and the targets in the moving state are behind the targets in the parking state, the speed of the targets in the moving state is greater than 10 km / h; or, when the number of targets in the parking state belongs to the interval [2, 3] and the targets in the moving state are in the surrounding lanes of the targets in the parking state, the speed of the targets in the moving state is greater than 20 km / h; the stability coefficient is used to characterize the possibility that the relevant event will continue to occur within a set time period; The off-peak hour prediction unit is configured to, for the frame scene corresponding to the off-peak time period, predict whether there will be a path conflict among the targets in the frame scene corresponding to the off-peak time period according to the state change data; The prediction result unit is configured to, if so, predict the time of the path conflict and obtain the high-definition surveillance video of the road traffic scene according to the time of the conflict; the resolution and frame rate of the high-definition surveillance video are both greater than those of the current surveillance video; The off-peak hour event detection unit is configured to perform event detection on the high-definition surveillance video by using the detection parameters corresponding to the off-peak time period to obtain a preliminary detection result.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the road traffic event analysis method according to any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by the processor, it implements the road traffic event analysis method according to any one of claims 1 to 4.

Images