A mean shift based small target tracking algorithm in colored video

It is difficult to track small targets in colored videos using traditional tracking algorithms because of the lack of the target image information. A tracking algorithm was proposed to track small targets in colored videos whose size was from 7×7 pixels to 25×25 pixels. The algorithm included five steps: (1) a bilinear interpolation algorithm was applied to enlarge the target and the surrounding region; (2) a histogram equalization algorithm was used to enhance the image features of the enlarged region; (3) the target model and the target candidates were established by kernel density estimation algorithm in the enlarged and equalized image space; (4) the target location in the enlarged image was obtained by mean shift method; and (5) the location of target was transformed from the enlarged image to the original image. The time complexity of the proposed algorithm is O(n). The testing 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 and 16 ms for the two testing cases in this study.