A target tracking method combining HOG and shape context

A target will change with the light, deformation and background, hence it is difficult to track the target. In particular, a target will be lost under the changing light and complex background when implementing the algorithm for tracking target based on a single feature. Therefore, HOG feature and shape context feature were combined effectively in this study. First, extract the HOG feature, dimensionality reduction was performed to the high-dimensional features of HOG to obtain a handful of useful information by the compressed sensing principle. Then, extract the shape context feature, normalized the dimension by linear interpolation to make its feature dimensions consistent with HOG (after sparse), and the weighted fusion of these two features was performed. Finally, Naive Bayesian Classifier was introduced to the tracking algorithm to estimate the target position effectively. Experimental result proved that the proposed algorithm could track the target stably in spite of the great variation of light and complex background.