Particle Filter Based Tracking for Crossing of Targets with Similar Pattern

Tracking is an important topic in computer vision and object recognition. Recently, a probabilistic approach using particle filters has been applied to track moving objects. This kind of approach often uses a color histogram to estimate a likelihood function for probabilistic tracking. When two similar objects cross each other in view, the likelihood becomes high for both. This often causes tracking to fail. This paper proposes a new method to address the object crossing problem. The method estimates the object region, splits the region into horizontal zones, and calculates similarity based on each split region and each horizontal zone of the target. The new method makes the tracking of similar targets more robust when those targets cross. Results are demonstrated on real video sequences.