Book

Description

The object-tracking technique uses a convex and monotonically decreasing kernel profile on color feature based histogram target representations to minimize the mean square error between the actual density and its estimate. Smoothness of the similarity function provides the efficient and fast target localization in the image space. The metric for the similarity function is derived from the Bhattacharyya coefficients, which is a function of normalized probability density functions of the target and target candidate. It gives a better indication of the target localization in image space when small changes occur in the target representation and the appearance. Mean shift procedure provides the root of the gradients and is a function of location, which eliminate the exhaustive search for the target and provide the new location in minimum iterations.