Enhancement of Mean Shift Tracking Through Joint Histogram of Color and Color Coherence Vector

Abstract

Tracking of an object in a scene, especially through visual appearance is weighing much relevance in the context of recent research trend. In this work, we are extending the one of the approaches through which visual features are erected to reveal the motion of the object in a captured video. One such strategy is a mean shift due to its unfussiness and sturdiness with respect to tracking functionality. Here we made an attempt to judiciously exploit the tracking potentiality of mean shift to provide elite solution for various applications such as object tracking. Subsequently, in view of proposing more robust strategy with large pixel grouping is possible through mean shift. The mean shift approach has utilized the neighborhood minima of a similarity measure through bhattacharyya coefficient (BC) between the kernel density estimate of the target model and candidate. However, similar capability is quite possible through color coherence vectors (CCV). The CCV are derived in addition to color histogram of target model and target candidate. Further, joint histogram of color model and CCV is added. Thus, the resultant histograms are empirically less sensitive to variance of background which is not ensured through traditional mean shift alone. Experimental results proved to be better and seen changes in tracking especially in similar color background. This work explores the contribution and paves the way for different applications to track object in varied dataset.