Feasibility study of pedestrian tracking from a moving Camera using a system model with motion information

Pedestrian recognition is required for automotive and surveillance applications. In such applications, accurate pedestrian recognition is indispensable and to achieve such recognition, recent schemes combine detection and tracking operation. Many pedestrian tracking schemes based on particle filter algorithm are proposed though they do not acquire sufficient accuracy under practical conditions. To achieve such accuracy using particle filter, a system model which represents the system dynamics appropriately must be applied. When this model is applied, it is possible to improve the accuracy without increasing the computational complexity. In this paper, a system model using motion information for pedestrian tracking that improves the accuracy using SIR particle filter from moving camera is proposed. Feasibility of the proposed system model is evaluated using on-board images of practical situation. As a result, proposed scheme achieved higher accurate and robust tracking compared to the conventional scheme.