Target tracking based on object's shape
I. V. Borisova, G. D. Yakovlev · 2016
In this paper we consider the problem of automatic tracking for targets in complex natural backgrounds. The background elements can mask the target for a while. Target has a low contrast relatively to the background, moreover, some background elements and target can have the same brightness. Target detection in each frame of the video sequence is proposed by using its shape. The target shape is formed by multilevel thresholding based on the between-class variance maximization. A special rule overwriting the standard so-called dynamic proximity measure is used to achieve stable tracking.