An improved object tracking algorithm based on image correlation

Abstract

Regarding common problems existing in current object tracking algorithms based on image matching, such as large computational cost, long matching time and difficulty to realize in real time, this paper provides an improved algorithm which is able to achieve an accurate and rapid tracking. The new algorithm firstly computes the object prediction position by constructing a novel Kalman predictor, and then an adaptive optimized matching is performed in a neighborhood of this prediction position so as to get the real object position rapidly. The experimental results show this algorithm is tractable and readily realizable. What's more important is that it decreases computational cost significantly but meanwhile shows better performance than traditional correlation-based tracking algorithms.