Target anomaly detection method, device and storage medium based on traffic surveillance video

Abstract

The present invention relates to the field of traffic monitoring video detection, in particular to a method, device and storage medium for detecting abnormality of a target based on traffic monitoring video, detecting a missing moving target and determining a candidate target ID; extracting a preset in the last image frame where the candidate target ID appears The moving target ID within the area is the suspected target ID, and the suspected collision event is determined by the distance between the suspected target ID and the equivalent line segment of the candidate target ID; the suspected collision event is determined according to the speed parameters corresponding to the suspected collision ID and the candidate target ID in the suspected collision event Whether the collision event is a traffic accident event; the embodiment of the present invention combines traffic accident science and computer vision technology to establish a rule model based on prior knowledge, which can accurately and quickly identify traffic accidents, and is applicable to all types of traffic participants and different The traffic flow state; and the embodiment of the present invention has compact logic and small amount of computation, and can meet the requirements of real-time operation.

Description

technical field [0001] The invention relates to the field of traffic monitoring video detection, in particular to a method, device and storage medium for detecting abnormality of a target based on traffic monitoring video. Background technique [0002] With the rapid development of intelligent transportation system, the research of traffic accident detection system based on computer vision technology and image processing technology has attracted extensive attention. There are various reasons for traffic accidents, and the types of traffic accidents are also very different due to the difference in vehicle types and accident severity. In different traffic environments, the occurrence of traffic accidents must be accompanied by abnormalities such as collisions between objects; when traffic accidents cause collisions between traffic participants, the speed and direction of the collision objects will change sharply; but normal traffic Under normal circumstances, the speed and di...

AI Technical Summary

Problems solved by technology

In different traffic environments, the occurrence of traffic accidents must be accompanied by abnormalities such as collisions between objects; when traffic accidents cause collisions between traffic participants, the speed and direction of the collision objects will change sharply; but normal traffic In some cases, when the object changes direction, the speed and direction of the object will also change. Therefore, due to the different severity of traffic accidents, it is difficult to distinguish between minor traffic accidents and normal direction change behaviors.

[0003] There are many ways to detect traffic accidents in the current technology, mainly including: one is to judge whether a traffic accident has occurred by analyzing the vehicle trajectory, specifically: first, perform trajectory clustering according to a series of vehicle trajectory, and similar The motion behavior is divided into one category; then a unified behavior model is established for the same type of motion behavior; finally, some trajectories obtained in real time are detected for abnormal trajectories. If some trajectories pass the detection, the driving behavior of the vehicle can be obtained according to the probability; the method The amount of calculation is low, and the real-time and confidence are low; the other is to judge whether a traffic accident occurs by monitoring abnormal changes in traffic flow

Specifically, without considering the driving speed and direction of each individual vehicle, regardless of the type of vehicle, and whether the moving target is a vehicle or a pedestrian, a macroscopic data statistics of the global traffic flow is made, and the traffic flow is monitored and analyzed in real time , so as to analyze and understand the impact of traffic accidents on traffic flow; this method is limited by different cities, different road sections and different time periods of the day, and the data of various basic parameters of traffic flow are very different

The above two methods have their own limitations, and cannot handle different types of traffic participants and traffic flow states at the same time

[0004] In summary, due to the complexity of the traffic environment, the current proposed traffic accident detection methods still have limitations in practical applications such as single types of traffic participants, limited traffic flow status, and poor real-time performance of the system.

Images