Tracking Humans in Real-Time by Opponent-Motion and Visual Attention

Abstract

This paper presents a biologically-inspired approach for human tracking by using a single feature of opponent-motion and visual attention mechanism. Opponent-motion is a significant and salient feature during human motion of walking, running, etc. By the interaction with the heuristic prior knowledge of human model, the obtained opponent-motion is used to derive motion saliency from its statistical computation for rapid selection of salient motion blobs of humans by visual attention. The biological motion with available opponent-motion, therefore, can be effectively detected and tracked in real-time. The experimental results have also shown that detecting and tracking humans by opponent-motion can greatly eliminate the effect of illumination change. This work demonstrated that the feature of opponent-motion can be used for real-time human detection and tracking in natural and clustered environments.