Road traffic tracking and parameter estimation based on visual information analysis using self-calibrated camera views

Traffic safety has drawn increasing interest for researchers lately. Monitoring road traffic, tracking vehicles on roads and analysis surrounding vehicle information from an ego vehicle can be used to help the improvement of traffic safety and also provide drivers with surrounding vehicle information and giving warning signs on risky situations. In this paper, we propose a novel road traffic tracking and analysis scheme based on visual information with self-calibrated camera views from a single camera. In the proposed method, vehicles entering the scene are tracked by using combination of detection and tracking. A reference model containing multiple instance of candidate vehicles as well as a negative set of background samples are used for the detection and tracking, where similarity of each candidate object is compared. To analyze the parameters related to dynamic vehicles, a novel method is proposed for camera self-calibration where a priori information is exploited. This enables mapping tracked vehicles in 2D images to the ground plane in the 3D coordinate system where parameters such as trajectory, speed, distance and density of vehicles may be accurately estimated. The main novelties of the proposed scheme lie in the combined tracking and detection, new formulae derived in camera selfcalibration, and the hybrid system that handles both daytime and nighttime scenarios. Experiments were performed on several videos containing two way traffic and road crossing, preliminary results and evaluation show that the proposed scheme is rather promising for such an application.