CN112509315B - Traffic accident detection method based on video analysis - Google Patents
Traffic accident detection method based on video analysis Download PDFInfo
- Publication number
- CN112509315B CN112509315B CN202011214279.5A CN202011214279A CN112509315B CN 112509315 B CN112509315 B CN 112509315B CN 202011214279 A CN202011214279 A CN 202011214279A CN 112509315 B CN112509315 B CN 112509315B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- parking
- information
- vehicles
- state
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
- G08G1/0175—Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/161—Decentralised systems, e.g. inter-vehicle communication
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention relates to a traffic accident detection method based on video analysis, which comprises the steps of obtaining a video stream of a target road section; identifying the video stream to generate a positioning frame corresponding to each vehicle in the video stream; preprocessing the positioning frame, and generating track information according to the positioning frame meeting preset conditions; the vehicle image includes a target vehicle; identifying trajectory information of a target vehicle; parking information of a suspected accident target; information on whether the suspected accident target meets or not; congestion status information of the scene; collecting accumulated time of parking and pedestrian information near the parking; and judging whether an accident occurs according to all the calculated information. The traffic accident detection method has the characteristics of high real-time performance and strong adaptability to environmental noise, and has high accuracy and low false alarm rate in actual detection.
Description
Technical Field
The invention belongs to the field of image technology identification, and particularly relates to a traffic accident detection method based on video analysis.
Background
The continuous improvement of the social and economic development level and the rapid development of science and technology bring a very rapid development speed to the automobile industry, and the number of motor vehicles is increasing day by day. The increase of the quantity of motor vehicles leads to the frequent occurrence of traffic accidents which directly threaten the life and property safety of people and have larger harmfulness. In order to reduce the influence and loss caused by traffic accidents as much as possible, it is necessary to control the occurrence rate of traffic accidents and improve the efficiency of traffic accident handling. At present, monitoring cameras are mostly installed on main roads and intersections of large, medium and small cities in China to monitor the roads in real time, the monitoring coverage rate is continuously increased, but at present, the manual operation is still used as the main way for finding and handling traffic accidents. The traffic accident detection system based on video analysis has important significance for improving the efficiency of the traffic police to arrive at the accident site and handling the traffic accident, and determines the life safety of accident injured personnel, the economic loss and the length of traffic jam time to a great extent.
The chinese patent publication No. CN105405297A, based on the optical flow field, analyzes the global traffic flow characteristics of the video frames and the local vehicle motion characteristics of the video frames to construct an accident detection model, and detects whether a traffic accident occurs. Because camera weak jitter and network pause phenomena exist in actual traffic video monitoring, dynamic indexes such as optical flow and vehicle speed utilized by the method can cause frequent sudden change, so that more false alarms are easily caused.
In recent years, deep learning methods are widely applied in the field of target detection, and although deep learning has outstanding capabilities in feature extraction and target classification, the method has the limitation that whether target detection objects in a training set are sufficient or not directly affects the final recognition effect, but traffic accident scenes generally do not have general features. Therefore, this kind of correlation works to some extent in identifying some specific kinds of accident collision scenarios under some specific scenarios, but still does not have strong generalization capability.
Disclosure of Invention
The invention aims to provide a traffic accident detection method based on video analysis, which utilizes traffic monitoring video stream data, can effectively detect and distinguish traffic accidents in a real-time scene, reduces the influence of environmental noise, improves the accident detection accuracy and efficiency, reduces the false alarm rate, and has higher robustness and practicability.
The technical scheme of the invention is as follows:
a traffic accident detection method based on video analysis comprises the following steps:
s1: acquiring a video stream of a target road section, and configuring a parking available area and a parking unavailable area in a video picture;
s2: carrying out vehicle identification on the video stream, and generating a positioning frame of a vehicle in the video stream;
s3: preprocessing the positioning frame, periodically detecting the positioning frame according to a preset time interval, acquiring track information of the positioning frame, wherein the track information comprises a vehicle ID (identity) in the positioning frame and positioning frame coordinate information for a plurality of times, and recording the track information to a historical information recording table;
s4: judging whether the vehicle is in a parking state, calculating parking information of the vehicle and recording the parking information to a parking information recording table;
s5: judging whether any two vehicles in the video stream meet, and recording meeting information to a meeting vehicle list;
s6: acquiring an accumulated value of parking time according to the parking information of the vehicle;
s7: calculating and recording pedestrian information near the vehicle in the parking state, and acquiring the staying time of the pedestrian;
s8: judging the accident type according to the type of the vehicle in the stop state, judging whether the accident occurs according to the meeting information, the accumulated value of the parking time and the pedestrian information, sending out early warning if the accident occurs, and clearing the cache if the accident does not occur.
Preferably, the model for identifying the vehicle in step S2 is a YOLO target detection model.
Preferably, the calculation of determining whether the vehicle is in a stopped state in step S4 includes the steps of: s4.1: according to the ID of the vehicle, calculating the intersection ratio between the current position of the positioning frame under the ID and the position of the positioning frame detected last time at a preset time interval, if the intersection ratio is greater than the set intersection ratio threshold, the number of times meets the set number threshold, determining that the vehicle corresponding to the ID is in a preliminary parking state, otherwise, determining that the vehicle is in a starting state;
s4.2: recording parking information of a vehicle in a preliminary parking state, calculating the parking distance of the current positioning frame position information of the vehicle in the preliminary parking state and the first-appearing position information of the positioning frame in the history information recording table corresponding to the vehicle ID, judging to slow down if the parking distance exceeds a set distance threshold value, eliminating false parking report, and otherwise, determining that the current vehicle is in the parking state.
Preferably, the parking information is recorded in a parking information list, the parking information is the number of times that the intersection ratio is greater than a threshold, and if a vehicle ID corresponding to the vehicle determined as the preliminary parking state already exists, the position information and the time information under the vehicle ID are updated; and inquiring whether the vehicle ID exists in the parking information list aiming at the vehicle ID corresponding to the vehicle which is determined to be in the starting state, and clearing the parking information under the vehicle ID if the vehicle ID exists.
Preferably, the specific step of determining whether the vehicles meet in step S5 is: and calculating the intersection ratio of the positioning frames between any two vehicles in the current scene aiming at the front frame and the rear frame of the video stream, if the intersection ratio of the positioning frames is larger than zero, the two vehicles meet, and recording the information of the meeting vehicles into a meeting vehicle list.
Preferably, the step of calculating the integrated value of the parking time in step S6 includes:
s6.1: judging whether the vehicles in the video stream are in a congestion state and a queuing state, if so, not accumulating the parking time of the vehicles;
s6.2: and accumulating the parking time of the vehicle in the parking state, if the intersection ratio of the positioning frames between the first occurrence and the last occurrence of the vehicle in the video stream is greater than a set threshold value, accumulating the parking time, and obtaining the accumulation of the parking time according to the time calculation time difference.
Preferably, the step of acquiring pedestrian information in step S7 is:
s7.1: separating the vehicle in the parking state and the pedestrian in the vicinity thereof;
s7.2: judging whether the pedestrian is in a circumscribed circle of the vehicle in a parking state;
s7.3: calculating and recording the staying time of the pedestrian in the 2 times of the circumscribed circle of the vehicle in the parking state;
preferably, the accident types include: motor vehicle and motor vehicle accidents, motor vehicle and non-motor vehicle accidents, and single vehicle accidents; and judging whether the parked vehicle is in a congestion state, if so, only judging the single vehicle.
Preferably, the judgment processes of the motor vehicle and motor vehicle accident and the motor vehicle and non-motor vehicle accident are the following steps: if the two vehicles meet each other, the staying time of the two vehicles exceeds 30 seconds, and the staying time of pedestrians nearby any one of the two vehicles to the vehicle within the range of 2 times of the circumscribed circle exceeds 30 seconds, the motor vehicle and motor vehicle accident is reported; if the two vehicles meet each other once and the staying time of the two vehicles exceeds 60 seconds, the motor vehicle and motor vehicle accident is reported; if the two vehicles do not meet, the accident judgment is not carried out.
Preferably, the process for judging the single-vehicle accident is as follows: the vehicle is a non-motor vehicle or a motor vehicle, and the stop time of the vehicle exceeds 150 seconds, and the single vehicle accident is reported.
Preferably, the method for determining whether the vehicle is in the congestion state includes: if the number of the stopped vehicles in the monitoring area is not less than the preset number threshold, counting the average vehicle speed in the monitoring area, wherein if the average vehicle speed is less than the preset vehicle speed threshold, the monitoring area is jammed, and then the vehicles in the video stream are judged to be in a jammed state.
Preferably, the method for judging whether the vehicle is in the queuing state comprises the following steps:
expanding the width of the positioning frame, and recording vehicles in the width range to a queuing list in the current scene;
sequencing the recorded vehicles according to the y coordinate of the center point of the positioning frame;
and judging whether the intersection ratio of the adjacent vehicles is greater than zero, and if the number of the adjacent vehicles meeting the conditions exceeds a preset threshold value, judging that the vehicles are in a queuing state.
The invention has the beneficial effects that:
1. the method provided by the invention has the advantages of high real-time performance and strong adaptability to environmental noise, and has high accuracy and low false alarm rate in actual detection.
2. The accident classification and judgment method provided by the invention summarizes a large number of actual scene accident scenes by investigation, provides a plurality of configuration schemes suitable for different accident scenes, and has popularization and practical values.
Drawings
FIG. 1 is a flowchart illustrating steps according to an embodiment of the present invention.
Detailed Description
The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the present invention provides a traffic accident detection method based on video analysis, which comprises the following steps:
1. and acquiring and processing a video stream, and dividing a parking area and an unavailable parking area.
As an embodiment of the invention, a camera is used for acquiring a video of a target road section, a parking available area and a parking unavailable area are configured in a video picture, in traffic forward checkpoint video monitoring, the parking available area is a parking area and a left-turning waiting area in front of a stop line, and the parking unavailable area is a vehicle driving area in a crossing; in the road section monitoring, the road areas in the video range are generally all non-parking areas; the method comprises the steps of setting corresponding parking alarm time for configured parking available areas and parking unavailable areas, wherein the alarm time is set according to different places and can be adjusted according to the field alarm condition.
2. And identifying the video stream, and generating a positioning frame of the vehicle in the video.
As an embodiment of the present invention, a YOLOv3 model is called to identify a video stream, and a positioning frame corresponding to each vehicle in the video stream is generated.
In the embodiment of the present invention, the output result of the YOLO model is:
{a1,a2,..,an},ai＝{xai,yai,wai,hai}；
wherein xai,yaiRespectively the abscissa and ordinate of the center point of the target positioning frame, wai,haiThe width and the height of the target positioning frame are respectively shown, a represents a detection result positioning frame and is a mark symbol, and a is { x, y, w, h }, and a data structure consisting of an abscissa, an ordinate, a width and a height represents a positioning frame.
3. And (4) preprocessing the positioning frame generated in the step (2) and acquiring track information.
As an embodiment of the present invention, an intersection ratio between the positioning frames in adjacent frames of the video stream is first calculated, and then track information is generated according to the positioning frame with the intersection ratio greater than an intersection ratio threshold.
An Intersection-over-unity (IOU), a concept used in target detection, is the overlap ratio of the generated candIDate frame (candIDate frame) and the original marker frame (ground round frame), i.e. the ratio of their Intersection to Union. Since the frames in the video stream that correspond to the same vehicle usually have a higher overlap, the intersection ratio of the frames in the video stream is calculated in this step, so as to determine the frames that correspond to the same vehicle.
Specifically, the intersection-to-parallel ratio IOU can be calculated according to the following formula:
area (a) and area (B) are the areas of the frame in the front and back frames of the video stream, respectively, and usually the frame in the front frame is a and the frame in the back frame is B.
The acquired track information comprises identification information of the vehicle and positioning coordinate information of the vehicle, and the track information is used for judging subsequent parking information.
In the embodiment of the invention, the track set queue structure is as follows:
TraceSet＝{trace1,trace2,...tracen}；
wherein TraceSet is a trace set queue, trace1Is the ith trackInformation, wherein the ID is a vehicle tracking number, namely identification information; TR is the coordinate of the central point of the bottom of the vehicle, wherein m is the number of positioning frames corresponding to the same tracking number; MATstart，MATendDividing the vehicle track into a starting point image and a finishing point image for judging whether the tracking is correct or not; wherein x, y and h are the abscissa, the ordinate and the height of the positioning frame.
4. And judging whether the video scene is in a congestion state.
In one embodiment of the present invention, when there are at least 5 vehicles in the monitored area, the average vehicle speed in the monitored area is counted, the average vehicle speed in 5 out of the last 10 counts is less than 0.01, and the shift speed of the pixel point position in the 1920 × 1080 video screen is defined as pixel/second, the unit of the vehicle speed in this embodiment is defined as pixel/second, and it is considered that the congestion occurs in the monitored area.
5. And judging whether the video scene is in a queuing state or not.
As an embodiment of the invention, the width of a positioning frame of any vehicle in a video is expanded by one fourth to the left and right, and vehicles within the width range in the current scene are calculated; and recorded into the queuing table; sorting the list according to the y coordinate of the central point; and traversing the list, judging whether the IOU of the adjacent vehicle is larger than zero, and if the condition is met, judging that the number of the adjacent vehicles exceeds three, and determining that the vehicles are in a queuing state.
6. And judging whether the vehicle is in a parking state or not, and calculating and recording parking information of the vehicle.
As an embodiment of the present invention, the steps are specifically as follows:
6.1: according to the vehicle ID, the position information and the time information of the last 20 times corresponding to each ID are stored in a history information recording table.
6.2: and calculating the detection results of the positioning frame position information of the current vehicle corresponding to each ID and the positioning frame at the previous set detection time interval moment according to the current vehicle information, and preliminarily judging that the vehicle stops if 5 times of the detection results meet the threshold value 0.9 of IOU parking.
6.3: the parking information of the vehicle in the parking state is stored in a parking information list, if the corresponding ID exists, the position information and the time information of the corresponding vehicle are updated, the distance between the corresponding vehicle and the first parking state of the corresponding vehicle is calculated, if the distance exceeds a preset slow-moving threshold value, the vehicle is judged to be in the slow-moving state, and the parking information of the ID is removed from the parking information list.
6.4: the parking information list is inquired whether or not there is an ID of the vehicle determined to be in the activated state, and if so, the parking information related to the ID is cleared.
7. And judging whether any two vehicles meet in the video stream.
As an embodiment of the present invention, the determining method in this step specifically includes:
7.1: and calculating the IOU of any two vehicles in the current scene aiming at the front video frame and the rear video frame.
7.2: if the IOU is larger than zero, the two vehicles meet, and the information of the meeting vehicles is recorded into a meeting vehicle list. The encountered vehicle information is a tuple (vehicle a, vehicle b) for marking the behavior that the two vehicles a and b have been close to each other and is used for one of the conditions for judging the vehicles with the suspected accident.
8. The parking accumulated time for stopping the vehicle is calculated.
8.1: firstly, recording parking information as StopList, recording historical recording information as prevList, and recording the latest 20 records of each vehicle ID by the prevList;
8.2: traversing the StopList, and accumulating the time of each stopped vehicle;
8.3: if the area of the vehicle is less than 1500 pixels, the vehicle is considered to be a distant vehicle, and time accumulation is not carried out;
8.4: if the position of the vehicle is judged to be in a queuing state, time accumulation is not carried out;
8.5: finding out a corresponding history record according to the tracking ID of the vehicle, if the IOU between the first occurrence and the last occurrence is less than 0.7, indicating that the vehicle moves and the parking time does not need to be accumulated; if the IOU is greater than 0.7, the tracking ID may have changed, requiring time to accumulate;
8.6 in the StopList, searching for a stopped vehicle with the IOU larger than 0.7, storing the stopped vehicle in a temporary list, and recording the stopped vehicle as tempList, wherein the tempList is sorted from large to small according to the size of the ID;
8.7: circularly traversing the tempList, then finding out historical record information from the prevList according to the ID, if the proportion of the historical record and the IOU of the vehicle, which is more than 0.8, exceeds 90%, recording time accumulation information, otherwise, not performing accumulation, and exiting the circulation;
8.8 for step 8.7, when traversing the tempList, determine whether the vehicle with the largest ID is coming from a distant place, if it is coming from a distant place, the loop exits if the ratio is greater than 90%.
9. Pedestrian information in the vicinity of the stopped vehicle is recorded.
As an embodiment of the present invention, the specific process of this step is:
9.1: separating out the stopped vehicle and the pedestrian in the scene; calculating pedestrian information around each stopped vehicle; storing a map, recording the map as stopPersonMap, and enabling each vehicle to correspond to a pedestrian information list;
9.2: taking a stopped vehicle C as an example, firstly searching whether corresponding information exists in stopPersonMap, if so, traversing the current pedestrian list, judging whether the pedestrian list corresponding to the vehicle C exists for each pedestrian, and if so, calculating the staying time of the pedestrian in a circumscribed circle 2 times of the vehicle C; if not, judging whether the pedestrian information is in the circumscribed circle of the vehicle C, and recording the pedestrian information in the circumscribed circle;
9.3: if the pedestrian information corresponding to the vehicle C is not found in the stopPersonMap, whether the pedestrian is in the circumscribed circle of the vehicle C needs to be judged.
10. And classifying the traffic accident type according to the obtained information.
Assuming that the time threshold is 30 seconds, taking the stopped vehicle D and the stopped vehicle E as an example;
if the current scene is judged to be in a congestion state, only judging the single-vehicle accident;
(1) motor vehicle and motor vehicle accident
1) The vehicles D and E are all motor vehicles, and the condition that the vehicles D and E do not meet is not judged;
2) if the vehicle D meets the vehicle E once, judging the accident, otherwise, not judging;
3) if the vehicle D meets the vehicle E once, the stop time of the two vehicles exceeds 30 seconds, and a certain pedestrian stays for 30 seconds in the range of 2 times of the circumscribed circle around the vehicle D or the vehicle E, the accident of the motor vehicle and the motor vehicle is reported;
4) and if the vehicle D meets the vehicle E once and the stopping time of the two vehicles exceeds 60 seconds, reporting the accident of the motor vehicle and the motor vehicle.
(2) Accidents of motor vehicles and non-motor vehicles
1) One of the vehicle D and the vehicle E is a motor vehicle, the other vehicle is a non-motor vehicle, and no judgment is made when the condition is not met;
2) if the vehicle D meets the vehicle E once, judging the accident, otherwise, not judging;
3) if the vehicle D meets the vehicle E once, the stop time of the two vehicles exceeds 30 seconds, and a certain pedestrian stays for 30 seconds in the range of 2 times of the circumscribed circle around the vehicle D or the vehicle E, the accident of the motor vehicle and the motor vehicle is reported;
4) and if the vehicles D and E meet once and the stopping time of the two vehicles exceeds 60 seconds, the motor vehicle and non-motor vehicle accidents are reported.
(3) Accident of bicycle
The vehicle is a motor vehicle or a non-motor vehicle, and the condition is not met without judgment; for example, if the stop time of the D vehicle exceeds 150 seconds, the single vehicle accident is reported.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present invention, which are used for illustrating the technical solutions of the present invention and not for limiting the same, and the protection scope of the present invention is not limited thereto, although the present invention is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the present invention in its spirit and scope. Are intended to be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (9)
1. A traffic accident detection method based on video analysis is characterized by comprising the following steps:
s1: acquiring a video stream of a target road section, and configuring a parking available area and a parking unavailable area in a video picture;
s2: carrying out vehicle identification on the video stream, and generating a positioning frame of a vehicle in the video stream; calling a YOLOv3 model in the step S2 to identify the video stream;
s3: preprocessing the positioning frame, periodically detecting the positioning frame according to a preset time interval, acquiring track information of the positioning frame, wherein the track information comprises a vehicle ID (identity) in the positioning frame and positioning frame coordinate information for a plurality of times, and recording the track information to a historical information recording table;
s4: judging whether the vehicle is in a parking state, calculating parking information of the vehicle and recording the parking information to a parking information recording table; the calculation of determining whether the vehicle is in a stopped state described in step S4 includes the steps of:
s4.1: according to the ID of the vehicle, calculating the intersection ratio between the current position of the positioning frame under the ID and the position of the positioning frame detected last time at a preset time interval, if the intersection ratio is greater than the set intersection ratio threshold, the number of times meets the set number threshold, determining that the vehicle corresponding to the ID is in a preliminary parking state, otherwise, determining that the vehicle is in a starting state;
s4.2: recording parking information of a vehicle in a preliminary parking state, calculating a parking distance between the current position of the vehicle in the preliminary parking state and the position, corresponding to the vehicle ID, of the vehicle in the parking state, where the vehicle enters the parking state for the first time in a parking information recording table, if the parking distance exceeds a set distance threshold value, determining to slow down, excluding false parking, and otherwise determining that the current vehicle is in the parking state;
s5: judging whether any two vehicles in the video stream meet, and recording meeting information to a meeting vehicle list;
s6: acquiring an accumulated value of parking time according to the parking information of the vehicle;
s7: calculating and recording pedestrian information near the vehicle in the parking state, and acquiring the staying time of the pedestrian;
s8: judging the accident type according to the type of the vehicle in the stop state, judging whether the accident occurs according to the meeting information, the accumulated value of the parking time and the pedestrian information, sending out early warning if the accident occurs, and clearing the cache if the accident does not occur.
2. The video analysis-based traffic accident detection method according to claim 1, wherein the parking information is recorded in a parking information list, the parking information is the number of times that the intersection ratio is greater than a threshold, and if a vehicle ID corresponding to a vehicle determined as a preliminary parking state already exists, the position information and time information under the vehicle ID are updated; and inquiring whether the vehicle ID exists in the parking information list aiming at the vehicle ID corresponding to the vehicle which is determined to be in the starting state, and clearing the parking information under the vehicle ID if the vehicle ID exists.
3. The video analysis-based traffic accident detection method according to claim 1, wherein the specific determination step of whether the vehicles meet in step S5 is: and calculating the intersection ratio of the positioning frames between any two vehicles in the current scene aiming at the front frame and the rear frame of the video stream, if the intersection ratio of the positioning frames is larger than zero, the two vehicles meet, and recording meeting information into a meeting vehicle list.
4. The video analysis-based traffic accident detection method according to claim 1, wherein the step of calculating the accumulated value of the parking time in step S6 is specifically as follows:
s6.1: judging whether the vehicles in the video stream are in a congestion state and a queuing state, if so, not accumulating the parking time of the vehicles;
s6.2: and accumulating the parking time of the vehicle in the parking state, and if the intersection ratio of the positioning frames between the first occurrence and the last occurrence of the vehicle in the video stream is greater than a set threshold, accumulating the parking time, wherein the accumulated value of the parking time is obtained by multiplying the number of times that the intersection ratio is greater than the threshold by a preset time interval.
5. The video analysis-based traffic accident detection method according to claim 1, wherein the step of acquiring pedestrian information in step S7 is:
s7.1: separating the vehicle in the parking state and the pedestrian in the vicinity thereof;
s7.2: judging whether the pedestrian is in a circumscribed circle of the vehicle in a parking state;
s7.3: and calculating and recording the staying time of the pedestrian in the 2 times of the circumscribed circle of the vehicle in the parking state.
6. The video analysis-based traffic accident detection method of claim 5, wherein the accident type comprises: motor vehicle and motor vehicle accidents, motor vehicle and non-motor vehicle accidents, and single vehicle accidents; and judging whether the parked vehicle is in a congestion state, if so, only judging the single vehicle.
7. The video analysis-based traffic accident detection method according to claim 6, wherein the motor vehicle-to-motor vehicle accident and the motor vehicle-to-non-motor vehicle accident are determined in a consistent manner, and are both: if the two vehicles meet each other, the staying time of the two vehicles exceeds 30 seconds, and the staying time of pedestrians nearby any one of the two vehicles to the vehicle within the range of 2 times of the circumscribed circle exceeds 30 seconds, the motor vehicle and motor vehicle accident is reported; if the two vehicles meet each other once and the staying time of the two vehicles exceeds 60 seconds, the motor vehicle and motor vehicle accident is reported; if the two vehicles do not meet, the accident judgment is not carried out;
the single vehicle accident judgment process comprises the following steps: the vehicle is a non-motor vehicle or a motor vehicle, and the stop time of the vehicle exceeds 150 seconds, and the single vehicle accident is reported.
8. The video analysis-based traffic accident detection method according to claim 4 or 6, wherein the method for determining whether the vehicle is in a congested state comprises: if the number of the stopped vehicles in the monitoring area is not less than the preset number threshold, counting the average vehicle speed in the monitoring area, wherein if the average vehicle speed is less than the preset vehicle speed threshold, the monitoring area is jammed, and then the vehicles in the video stream are judged to be in a jammed state.
9. The video analysis-based traffic accident detection method according to claim 4, wherein the judgment method of whether the vehicle is in the queue state is:
expanding the width of the positioning frame, and recording vehicles in the width range to a queuing list in the current scene;
sequencing the recorded vehicles according to the y coordinate of the center point of the positioning frame;
and judging whether the intersection ratio of the adjacent vehicles is greater than zero, and if the number of the adjacent vehicles meeting the conditions exceeds a preset threshold value, judging that the vehicles are in a queuing state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011214279.5A CN112509315B (en) | 2020-11-04 | 2020-11-04 | Traffic accident detection method based on video analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011214279.5A CN112509315B (en) | 2020-11-04 | 2020-11-04 | Traffic accident detection method based on video analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112509315A CN112509315A (en) | 2021-03-16 |
| CN112509315B true CN112509315B (en) | 2022-02-15 |
Family
ID=74955211
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011214279.5A Active CN112509315B (en) | 2020-11-04 | 2020-11-04 | Traffic accident detection method based on video analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112509315B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113158835A (en) * | 2021-03-31 | 2021-07-23 | 华南理工大学 | Traffic accident intelligent detection method based on deep learning |
| CN113792586A (en) * | 2021-08-04 | 2021-12-14 | 武汉市公安局交通管理局 | Vehicle accident detection method and device and electronic equipment |
| CN113947907A (en) * | 2021-10-28 | 2022-01-18 | 高新兴科技集团股份有限公司 | Vehicle traffic accident early warning method, device, medium and equipment based on V2X |
| CN114023069B (en) * | 2021-11-18 | 2022-07-29 | 北京图盟科技有限公司 | Abnormal parking detection method, device and equipment and readable storage medium |
| CN114596508A (en) * | 2022-03-14 | 2022-06-07 | 浙江这里飞科技有限公司 | Violation identification method and device based on unmanned aerial vehicle, computer equipment and storage medium |
| CN115049993B (en) * | 2022-08-17 | 2022-11-15 | 成都考拉悠然科技有限公司 | Vehicle abnormal stop monitoring method based on deep learning |
| CN117292338B (en) * | 2023-11-27 | 2024-02-13 | 山东远东保险公估有限公司 | Vehicle accident identification and analysis method based on video stream analysis |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104575043A (en) * | 2014-11-28 | 2015-04-29 | 北京尚易德科技有限公司 | Automatic prompt system and method during passing of motor vehicle through pedestrian crosswalk |
| CN104680133A (en) * | 2015-02-04 | 2015-06-03 | 上海依图网络科技有限公司 | Real-time detecting method for pedestrian avoidance behavior of illegal vehicle |
| CN105513342A (en) * | 2015-11-25 | 2016-04-20 | 南京莱斯信息技术股份有限公司 | Video-tracking-based vehicle queuing length calculating method |
| CN106355919A (en) * | 2016-09-21 | 2017-01-25 | 山西省交通科学研究院 | Machine-vision-based curve speed limitation calculating method and speeding warning device |
| CN109747642A (en) * | 2019-01-04 | 2019-05-14 | 北京博宇通达科技有限公司 | Vehicle travel control method and equipment |
| CN110084308A (en) * | 2019-04-30 | 2019-08-02 | 杭州远眺科技有限公司 | A method of building commuting path is analyzed based on track of vehicle |
| WO2020024457A1 (en) * | 2018-08-01 | 2020-02-06 | 平安科技（深圳）有限公司 | Liability cognizance method and device of traffic accident and computer readable storage medium |
| CN111800507A (en) * | 2020-07-06 | 2020-10-20 | 湖北经济学院 | Traffic monitoring method and traffic monitoring system |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101288312B1 (en) * | 2013-03-05 | 2013-08-07 | 전재현 | Auto inspection attached system converged image recognition camera and the method thereof |
| US10336318B2 (en) * | 2015-06-22 | 2019-07-02 | Ford Global Technologies, Llc | Systems and methods for vehicle park assist |
| CN105761500B (en) * | 2016-05-10 | 2019-02-22 | 腾讯科技（深圳）有限公司 | Traffic accident treatment method and traffic accident treatment device |
| KR101688218B1 (en) * | 2016-08-25 | 2016-12-20 | 델리아이 주식회사 | System for managing traffic flow and incidents using image detection in real time based on object recognition and method for processing thereof |
| CN106652465B (en) * | 2016-11-15 | 2020-04-07 | 成都通甲优博科技有限责任公司 | Method and system for identifying abnormal driving behaviors on road |
| CN107730903A (en) * | 2017-06-13 | 2018-02-23 | 银江股份有限公司 | Parking offense and the car vision detection system that casts anchor based on depth convolutional neural networks |
| CN107491753A (en) * | 2017-08-16 | 2017-12-19 | 电子科技大学 | A kind of parking offense detection method based on background modeling |
| DE102017218921B4 (en) * | 2017-10-24 | 2024-05-23 | Bayerische Motoren Werke Aktiengesellschaft | Method, device, computer program and computer program product for operating a display unit of a vehicle |
| CN109727456A (en) * | 2017-10-30 | 2019-05-07 | 杭州海康威视数字技术股份有限公司 | A kind of method and device identifying vehicles identifications |
| CN109800696A (en) * | 2019-01-09 | 2019-05-24 | 深圳中兴网信科技有限公司 | Monitoring method, system and the computer readable storage medium of target vehicle |
| CN111369807B (en) * | 2020-03-24 | 2022-04-12 | 北京百度网讯科技有限公司 | Traffic accident detection method, device, equipment and medium |
| CN111626275B (en) * | 2020-07-30 | 2020-11-10 | 江苏金智慧安科技有限公司 | Abnormal parking detection method based on intelligent video analysis |
| CN111626277B (en) * | 2020-08-03 | 2021-02-26 | 杭州智诚惠通科技有限公司 | Vehicle tracking method and device based on over-station inter-modulation index analysis |
-
2020
- 2020-11-04 CN CN202011214279.5A patent/CN112509315B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104575043A (en) * | 2014-11-28 | 2015-04-29 | 北京尚易德科技有限公司 | Automatic prompt system and method during passing of motor vehicle through pedestrian crosswalk |
| CN104680133A (en) * | 2015-02-04 | 2015-06-03 | 上海依图网络科技有限公司 | Real-time detecting method for pedestrian avoidance behavior of illegal vehicle |
| CN105513342A (en) * | 2015-11-25 | 2016-04-20 | 南京莱斯信息技术股份有限公司 | Video-tracking-based vehicle queuing length calculating method |
| CN106355919A (en) * | 2016-09-21 | 2017-01-25 | 山西省交通科学研究院 | Machine-vision-based curve speed limitation calculating method and speeding warning device |
| WO2020024457A1 (en) * | 2018-08-01 | 2020-02-06 | 平安科技（深圳）有限公司 | Liability cognizance method and device of traffic accident and computer readable storage medium |
| CN109747642A (en) * | 2019-01-04 | 2019-05-14 | 北京博宇通达科技有限公司 | Vehicle travel control method and equipment |
| CN110084308A (en) * | 2019-04-30 | 2019-08-02 | 杭州远眺科技有限公司 | A method of building commuting path is analyzed based on track of vehicle |
| CN111800507A (en) * | 2020-07-06 | 2020-10-20 | 湖北经济学院 | Traffic monitoring method and traffic monitoring system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112509315A (en) | 2021-03-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112509315B (en) | Traffic accident detection method based on video analysis | |
| US6442474B1 (en) | Vision-based method and apparatus for monitoring vehicular traffic events | |
| Ki et al. | A traffic accident recording and reporting model at intersections | |
| KR101095528B1 (en) | An outomatic sensing system for traffic accident and method thereof | |
| CN106297278B (en) | Method and system for querying a projectile vehicle | |
| CN109326124A (en) | A kind of urban environment based on machine vision parks cars Activity recognition system | |
| CN113676702B (en) | Video stream-based target tracking and monitoring method, system, device and storage medium | |
| CN110796862B (en) | Highway traffic condition detection system and method based on artificial intelligence | |
| CN113936465B (en) | Traffic event detection method and device | |
| CN113012436B (en) | Road monitoring method and device and electronic equipment | |
| CN111292353A (en) | Parking state change identification method | |
| KR101133222B1 (en) | System and Method for monitoring traffic and accident based on image processing | |
| CN114648748A (en) | Motor vehicle illegal parking intelligent identification method and system based on deep learning | |
| KR101828271B1 (en) | Traffic management system and method thereof | |
| US20210397187A1 (en) | Method and system for operating a mobile robot | |
| CN111105619A (en) | Method and device for judging road side reverse parking | |
| CN113870551A (en) | Roadside monitoring system capable of identifying dangerous and non-dangerous driving behaviors | |
| Kamijo et al. | Development and evaluation of real-time video surveillance system on highway based on semantic hierarchy and decision surface | |
| CN111291722A (en) | Vehicle weight recognition system based on V2I technology | |
| CN116311166A (en) | Traffic obstacle recognition method and device and electronic equipment | |
| CN114882709B (en) | Vehicle congestion detection method, device and computer storage medium | |
| CN110782485A (en) | Vehicle lane change detection method and device | |
| CN113223276B (en) | Pedestrian hurdling behavior alarm method and device based on video identification | |
| Al-Kadi et al. | Road scene analysis for determination of road traffic density | |
| Tsai et al. | Multi-lane detection and road traffic congestion classification for intelligent transportation system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Xia Yingjie Inventor after: Qian Nan Inventor after: Zhang Lei Inventor before: Xia Yingjie Inventor before: Ma Oubo Inventor before: Liu Xuejiao |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |