Visual tracking based on improved foreground detection and perceptual hashing

The two well-known problems of the existing inter-frame differencing are foreground aperture and ghosting for low-speed moving targets or homogeneous color targets. To alleviate these problems, in this paper, we first propose an improved three-frame differencing algorithm and a foreground detection method combining background subtraction with the improved three-frame differencing. We calculate the average detection error rate of these three algorithms which are inter-frame differencing, improved three-frame differencing, and the foreground detection method combining background subtraction with the improved three-frame differencing. Experimental results demonstrated that our proposed methods ensure fast and accurate foreground detection. Moreover, considering that there are few studies applying perceptual hashing to object tracking, a new object tracking approach using perceptual hashing algorithms is presented. Perceptual hashing for object tracking is easy to achieve and the experimental results indicate that aHash, pHash and dHash all can track moving objects effectively and efficiently. Besides, we perform a comparison of three basic hash tracking algorithms and a comparison of the computational cost of different tracking window sizes for the same tracking algorithm. We intend to use them in different occasions for future applications, such as face tracking, robot localization and navigation, infrared thermal imaging tracking.