Perceptual object tracking

This paper presents an improved kernel-based target tracking that uses new and effective features able to describe the target appearance. The key idea consists of adopting features that are related to the visual perception of the target in place of its color histogram. The change of the feature space is twofold advantageous. It allows us to faithfully follow the target and to considerably reduce the computational cost of the whole tracking algorithm thanks to the reduced dimension of the perceptual feature space. Preliminary tests on some video sequences are encouraging. The proposed tracker is able to follow the target even in case of complete occlusion for some consecutive frames. Moreover, it is more stable than the algorithm that uses just the luminance histogram as target feature. Finally, it is robust to the presence of other moving targets.