High-quality vehicle trajectory generation from video data based on vehicle detection and description

Vehicle trajectories contain rich information on microscopic phenomena such as car following and lane changing. Despite many efforts to retrieve reliable trajectories from video images, previous approaches do not give high enough quality of trajectories that can be used in microscopic analysis. We introduce a new vehicle tracking approach based on a model-based 3-D vehicle detection and description algorithm. The proposed algorithm uses a probabilistic line feature grouping method to detect vehicles with little computation. A dynamic programming algorithm is proposed for fast reasoning. We present the system implementation and the vehicle detection and tracking results.