Infrared object tracking based on adaptive multi-features integration

Designing on effective observation model to discriminate object region from complex background is the core of robust tracking. A tracking approach based on multi-features observation has been proposed for infrared image sequences. Object appearance is represented by gray value, local standard deviation and gradient features in a unified histogram form; a scence-adaptive weighting scheme for these three features is used to construct the observation model, the selection of these multifeatures weights is towards the direction of maximizing discriminability between the target and its adjacent background. Experimental results on real complex situation demonstrate that the proposed algorithm tracks target well in highly appearance changes and severe clutter.