Real-time detection of passing objects using virtual gate and motion vector analysis

Abstract

Real-time human tracking and pedestrian counting in very complex situations with different directions of motion has been important for video surveillance and our daily life applications. This work presents a virtual gate method for the pedestrian detection without the need to construct a background model a priori. The proposed method utilizes motion estimation with three step search and a novel motion vector analysis algorithm which detects moving objects passing through the gate along any desired direction. This method is particularly applicable to complex situations. The experimental results demonstrate that the proposed strategy is reliable. © 2008 Springer-Verlag Berlin Heidelberg.