Hamming Distance Based Visual Tracking in Image Sequences

Abstract

Hamming distance (HD) based gradient orientation pattern matching (GOPM) has been proven to be a robust object tracking technique against varying illuminations and occlusions in an image. The technique is baased on the template matching technique with unit gradient vectors (UGVs) instead of traditional image features such as intensities and gradients. This is due to the fact that UGVs are insensitive to illumination changes. Furthermore, to cope with the so-called occlusion problem, we search for a target pattern using the HD metric. The traditional HD metric, however, measures the similarity between patterns by the logical exclusive OR (XOR) operation and is suitable for binary encoded information only. This paper is concerned with the conversions of the similarities between UGVs to binary signals in order that the HD can be applied. We also investigate the effect of the threshold values used in the conversion.

Keywords: object tracking, gradient orientation pattern matching, Hamming distance, unit gradient vectors