Human Tracking Based on Particle Filter in Outdoor Scene

In this paper, we propose the object tracking method based on color histograms and particle filtering. Particle filtering is a time series filter for estimating a state using probabilistic approach. Unlike deterministic approach such as template matching algorithm, it is more robust to occlusion or clutter because of its having many hypotheses. Moreover, color histograms are robust to partial occlusion, scale invariant and computational efficient. However, histogram has no spatial information. To solve this, multi-part histogram which has color histograms divided into sub-regions has been proposed in the past. However, multi-part histogram also has a disadvantage of no working in tracking an object having plain color. Therefore, we propose the adaptive target color representation by determining to use which multi-part color histogram or single-part color histogram in initializing a target. Our tracking method successfully tracks an object having multi-color or plain color on video sequences in outdoor scene.