A novel target tracking algorithm under complex background
Jingyu Cao, Weiyan Chai, Peizhi Liu · 2011
In order to realize the stable, accurate and realtime target tracking, a novel target tracking algorithm combining scale-invariant features extraction and fast pattern matching is proposed. First, create Gaussian Pyramid for decomposing the target template and decomposing the current image under test, extract the scale-invariant features for both the target template and the current image under test, locate the target in the image based on the corresponding scale-invariant features, and update the template. Second, calculate the normalized cross correlation between the decomposed target template and the decomposed current image under test in the frequency domain; select the Region of Interest (ROI) based on the maximum correlation value in the high resolution image for accurate target tracking. During the process, update the template in time to ensure the stable tracking when the target changes. Experimental results demonstrate that this proposed algorithm is superior to traditional target tracking algorithm in stability, accuracy and real-time performance.