Visual attention & multi-cue fusion based human motion tracking method

A novel visual tracking algorithm based on visual attention and multiple cues fusion for human motion analysis is proposed in this paper. An auxiliary object is selected through visual attention mechanism. Feature of target position prediction and feature of motion continuity and color feature are used to determine the location of a target and an auxiliary object. Candidate color feature includes feature of hue and saturation, features of R (Red) channel, G (Green) channel, B (Blue) channel and linear combination of R, G and B when a target is tracked. Candidate color features with high reliability are dynamically selected and weight values of the color features are dynamically adjusted according to the change of a scene. Candidate color feature includes feature of hue and saturation when an auxiliary object is tracked. Combining with CAMSHIFT (Continuously Adaptive Mean Shift) technique, experimental results show that this new algorithm is more robust than the traditional static non-adaptive algorithm and gets better tracking effect than CMET (Collaborative Mean Shift Tracking). It can handle the situation that a target is occluded and that a target and its background color are similar. It can also track a cross-border target.