A method of low-rank decomposition with feature point detection for moving target tracking

Abstract

In respect of the problems of complex changing of targets and background in moving target tracking, we utilize feature point detection method under low-rank sparse decomposition to track the moving target. First, for video sequence image, we adopt RPCA (Robust Principal Component Analysis) algorithm to conduct low-rank sparse decomposition, thereby extracting the foreground of the target; then SURF operator is used on each frame of image in the foreground sequence to conduct feature point detection, which just operates on the foreground of the image, thereby reducing the disturbance of complex background for target tracking and meanwhile reducing the amount of calculation in feature point detection. Through experiment on scale change, non-rigid motion, partial occlusion, and low-resolution thermal imaging video sequence as well as comparison with such moving target tracking methods as MIL, L1APG and DFT, the validity of the method adopted in this article in tracking moving target is verified.