A Scale Adaptive Tracker Using Hybrid Color Histogram Matching Scheme

TLDR: An important property of the Chi-squared measure is reported- It outperforms Bhattacharyya measure in the task of histogram matching from a few significantly similar multimodal histograms.

Abstract: In this paper we propose a computationally efficient scale adaptive tracking method using a hybrid color histogram matching scheme. Firstly, we report an important property of the Chi-squared measure- It outperforms Bhattacharyya measure in the task of histogram matching from a few significantly similar multimodal histograms. Also, Bhattacharyya measure performs better while selecting matches from a varied dataset. We employ these results to develop a hybrid histogram matching procedure using the two measures. This method is used for a patch matching algorithm in real time tracking. We first calculate a color histogram of the target which is then compared with histograms of patches in the neighborhood in subsequent frames using this hybrid procedure to obtain the best match. We devise a systematic scale adaptive tracking method which is robust to rapid changes in the target size. It is also robust to partial occlusion of the target. Extensive experimental proof based on real life and test datasets is presented which demonstrates the excellent tracking accuracy achieved by our algorithm at real time.