Adaptive thresholding using particle filter for tracking small and low contrast objects

In this paper, we present a simple and robust method for tracking small and low contrast objects in video sequences. The technique is based on image segmentation by adaptive thresholding, which is done using a particle filter. In order to achieve this, the threshold is made a state of the system dynamics. Prior knowledge of the target attributes such as position, size and mean intensity are incorporated into the tracking algorithm. This novel idea resolves many challenging issues faced by most of the tracking algorithms e.g. sudden illumination changes, unpredictable motion and incorrect model update in consecutive frames.