Combining hausdorff distance, HSV histogram and nonextensive entropy for object tracking
Paulo Sergio Rodrigues, Gilson Antonio Giraldi, Jasjit S. Suri · 2006
Abstract. In Computational Vision, object tracking in a sequence of frames is one of the most important problems. Among the most used approaches there is the model-target one, which matches a model object against a candidate target region in a frame sequence. To accomplish this task, the Hausdorff distance has an attractiveness due to its simplicity of implementation and possibility of matching between two sets with different cardinal-ity. Viewing images as non-extensive systems, we may apply the Tsallis Entropy (which works with only one parameter, called entropic parameter) to segment the frames in order to find the target object. In this work we propose a methodology which combines Haus-dorff distance, Bayesian network, HSV histogram and Tsallis non-extensive entropy for objects recognition and tracking in a frame sequence. With this proposal, we reduce the Hausdorff’s noise sensitive and the high parameter dependence of the tracking task. We apply our method in experiments with one object over a moving background in a sequence of 300 frames. 1