Research On Vehicle Tracking Algorithm Based On Image Sequences

Video target tracking is a research hot spot in the field of computer vision and one of the core technologies in intelligent traffic system. As an effective means to get traffic information,it’s of great significance to have further study on video target tracking.The main research content of this paper is video-based target tracking algorithm. The paper mainly focuses on two different target tracking schemes to obtain the stable tracking trajectories. The works are as follows:1. The 2D tracking algorithm: the tracking algorithm, which relies on feature-points,solves tracking the target vehicles in 2D image space, including feature-points extraction,feature-template search matching, similarity measurement and template renewing method etc.When ignoring the feature-point’s height, the algorithm can get the feature-point’s projection speed corresponding to the road plane.2. The 3D tracking algorithm: the algorithm includes the inverse-projection data structure and 3D tracking algorithm implementation. According to the imaging principle of camera, we can establish the mapping relationship between the inverse-projection plane and the image plane. The inverse-projection image is the accurate reconstruction of target local surface. As the inherent properties of the target, a stable tracking can be achieved in the 3D real space.The inverse-projection image is a special form of image. There are some similarities and differences between the 2D tracking algorithm and the 3D tracking algorithm. The paper provides a detailed study of the tracking frame design and template search matching in inverse-projection images. Inverse-projection plane corresponds to the target’s actual measurement in 3D space. According to this algorithm, the accurate speed of the vehicle can be acquired.3. Application of Kalman Filter in target tracking algorithm. The 3D tracking algorithm avoids the nonlinear motion during linear moving target in 2D image space. Kalman Filter can help to estimate the motion position of the target and improve the tracking accuracy.Though experiments, it proves that these two algorithms are feasible. Using the different speeds obtained by the two tracking algorithms, we easily can also get 3D coordinates of thefeature-points and other information.