An Automatic Recognition Approach for Traffic Congestion States Based on Traffic Video

Abstract

With the increasing demand for traffic information services as well as the extensive deployment of traffic video surveillance, there is a critical need for realizing automatic identification of congestion state with traffic video. To this end, this study proposes a traffic congestion evaluation model with adaptive learning ability. The qualitative process of the proposed model has been previously analyzed. In this method, the video image feature sets are extracted initially, followed by the state classification model training and learning via support vector machine. Subsequently, genetic algorithm is used to realize the online adaptive optimization. The field experimental results indicate that this method has high recognition accuracy, fast processing speed, and strong adaptive ability, and it can provide an appropriate solution for solving the problem of all—day traffic congestion states recognition based on the traffic video information.