The Vehicle Detection And Vehicle Behaviour Recognition Based On Traffic Video

Traffic video contains the most comprehensive traffic information, and can be used to infer and understand the contents of the traffic behavior and accident. Highway traffic video monitoring system plays a more and more important role in intelligent transportation system. The research of vehicle detection and vehicle recognition based on traffic video has many key problems which need to be solved, such as the moving vehicle can’t extract effectively and the accuracy rate of vehicle behavior recognition still low.In order to identify the behavior of the highway vehicle effectively, this paper studies some related researches of the key technology about vehicle detection based on motion information, vehicle detection based on feature information, the low accuracy rate of vehicle behavior recognition and so on. In this paper, the main contents include as follows:(1)According to the requirement of vehicle detection in real time on highway, vehicle detection based on motion information is studied in this paper. In order to solve the problem about the emerge of trailing in moving vehicle detection while using the traditional Mixed Gauss Model, a modeling method by combining mixed Gauss model and adaptive background selective learning model is proposed, which makes the vehicle target extraction more complete and accurate. In addition, in order to improve the adaptability of the ViBe(Visual Background Extractor) model in different scenarios, this paper uses cone model to replace the Euclidean distance to calculate the distance of model, and calculate the sample variance model to determine the distance threshold adaptively.(2) Only using motion information is unable to confirm whether the target is a vehicle, this paper does research about vehicle detection based on feature information. According to different apparent characteristics of front and rear vehicle, this paper proposes a vehicle recognition method based on the combination of Haar-like and HOG feature in traffic video to improve the accuracy; However, the speed of vehicle detection based on feature information is slow, this paper combines motion and feature information, then proposes a moving vehicle detection method based on region of interest and feature information to narrow the search range of vehicles,and speed up the detection.(3) Due to a small amount of abnormal trajectory, such as lane changing and overtaking, the classic spectral clustering with longest common sub-sequence(LCSS) can’t effectively distinguish all kinds of trajectory. In addition, the popular HMM trajectory model ignores the negative impact of the samples and only classifies by maximum likelihood value to cause a higher rate of false recognition in vehicle behavior recognition. This paper proposes a vehicle trajectory recognition method based on quadratic spectral clustering and HMM-RF hybrid model, which can effectively distinguish overtaking, changing lanes and normal trajectory, and improve the accuracy rate and robustness of behavior recognition.(4) Based on the study of vehicle detection and behavior recognition method in this paper, the paper designs and implements the prototype system of vehicle recognition based on traffic video, which can recognize the vehicle behavior on the highway automatically.