Camshift has been well accepted as one of the most popular methods for object tracking. However, it fails to address complex situations, such as similar color interference, object occlusion, and illumination changes. In this paper, we enhance Camshift to enable it to handle the above-mentioned problems. A two-dimensional (2D) histogram of the hue and luminance is used for the color features of the target. To reduce the influence from irrelevant background pixels, a Flood-fill operation is implemented. The obtained 2D target model can precisely describe the target as well as achromatic points. A similarity score is evaluated to prevent similar color interference. When a target’s colors are not distinguishable from the background colors, motion information will contribute to the tracking task. Finally, an average rate change is adopted to maintain progressive but not abrupt changes in the window size. The proposed algorithm has been extensively tested. The results are satisfactory while maintaining the processing in real time.
Multimedia Tools and Applications 74(23), pp.10291–10312