Tracking Object by Combining Particle Filters and SIFT Features
Bin Feng, Bing Zeng, Jinbo Qiu · 2009
An object tracking algorithm based on the particle filter framework is proposed in this paper for video surveillance applications. The color histogram is combined with a scale invariant feature transform (SIFT) descriptor to represent the likelihood between the candidates and observed objects. They are then incorporated into the particle filter based tracking algorithm in order to achieve more robust and accurate results. As a complement to the similarity in color distribution, a measure of texture similarity is computed in terms of the number of matched SIFT key-points and the distance between two matched points. The use of a SIFT descriptor can enhance the tolerance to scale change, appearance variation, partial occlusion and cluttered background which are very popular in video surveillance practice. Experiments are provided to demonstrate that the performance of the proposed algorithm outperforms that of the traditional particle filter that makes use of the color histogram only.