Video target tracking based on fusion state estimation

Abstract

In this paper, a new fusion state estimation method by fusing extended Kalman filter with particle filter is proposed to realize efficient and robust video target tracking. Extended Kalman filter has the time performance close to the Kalman filter and is more suitable for nonlinear video target tracking. Particle filter is based on non-parameter estimation and outperforms in robustness in video tracking. Fusion state estimation can obtain more accurate and reliable motion state of video target by optimizing the state estimation and prediction of video target. To further boost the efficiency of video tracking, this paper also presents an adaptive frames sampling method which utilizes the motion state of video target to skip some frames and then avoid frame by frame sampling. In addition, an efficient video target state observation method is introduced. This method integrates adaptive background updating, adjacent three frames difference and canny edge detection to efficiently obtain the target contour and normalized HSV color histogram which are both crucial for video target matching.