Robust tracking under complex environments

A subspace tracking method is proposed to track targets under complex environments. First, target is represented by subspace histogram of oriented gradient descriptor, which is computed by projecting histograms of oriented gradient descriptor to a subspace, that is built from a large set of training samples before tracking. Then an integral histogram method is incorporated to reduce the computational complexity. By these, the tracking problem is formulated as a state inference problem under the particle filter framework of target motion parameter estimation. Numerous experiments demonstrate the effectiveness of our proposed method in indoor and outdoor environments where the targets are subject to blurring, pose, scale, and illumination changes as well as various other noises. Compared with the traditional template based trackers, our proposed tracking method manifests good performance in the ground reconnaissance application.