A Compressive Tracking Method Based on Gaussian Differential Graph and Weighted Cosine Similarity Metric

This letter presents an extended compressive tracking algorithm to increase the stabilization and robustness in scale variation, rotation, and illumination variation. The features are extracted from the Gaussian differential graphs and taken as input signals of compressive sensing. In order to reduce the computational cost, the operating area is narrowed down to the region of interest. The algorithm utilizes a naïve Bayes classifier to get N candidate target locations with N highest classifying scores. Their weighted multiframe cosine similarities with the ground truth object from the initial frame and the most similar object in some adjacent frame are calculated to find the target location in current frame. Experimental results demonstrate the superiority of the proposed algorithm over some state-of-the-art tracking algorithms. It also can efficiently meet the needs of real-time tracking.