Gaussian background mixture model based automatic incident detection system for real-time tracking

摘要

The image-based incident detection system is capable of not only replacing the loop detector, which has limited management and operation functions, but can also record the sequential conditions before and after a traffic accident. Thus, it is possible to analyze its mechanisms objectively using this data. In this study, the researchers developed a reliable video image based accident detection system with a high detection rate and low error rates. The proposed accident detection algorithm in this study provides the preliminary judgment of potential accident by detecting stopped objects using the Gaussian Mixture Model. Afterwards, it measures the traces of vehicles, the speed variance, and the occupancy per detecting area to propose an algorithm that makes the final accident decision. The proposed algorithm performs accident detection by extracting the stopped objects based on the video image, the trajectory movement of the trailing vehicle, and the variance of speed and traffic volume in the detection area. Thus, it can minimize false detections and maximize the detection rate, making it possible to accurately interpret an accident site and the circumstances surrounding it. Moreover, it is advantageous that the detection rate does not decline under bad weather conditions such as cloudy, rainy, foggy, or snowy.