A Bilinear Interpolation Mean Shift Small Target Tracking Algorithm

Yimei Kang,Guan Wang,Jiang Hu

DOI: https://doi.org/10.1109/ICSPCS.2011.6140873

2011-01-01

Abstract:It is difficult to track targets in grayscale videos especially for small targets because of the lack of the target image information. A tracking algorithm was proposed to track small targets in grayscale videos whose size was from 7x7 pixels to 25x25 pixels. The algorithm included six steps: (1) enlarge the target and the surrounding region by bilinear interpolation; (2) enhance the target features of the enlarged region by histogram equalization; (3) expand the single grayscale video into three channel video by pixel gradient; (4) establish the target model and the target candidates by kernel density estimation in the enlarged and equalized image space; (5) obtain the target location in the enlarged image by mean shift method; and (6) transform the location of target from the enlarged image to the original image. The time complexity of the proposed algorithm is O(n). The experimental results showed that the algorithm was able to track the small targets steadily, accurately and quickly. The deviation of target location was zero for most frames and no more than 2 pixels for a few frames in which the target rotated at a large angle. The time spent in tracking the small target in a frame was 15 or 16 ms for the two testing cases in this study.