Coast mode is one of tracking modes that deals with the target occlusion, where tracking is halted for a while and the servo slew rate is maintained according to the target movement just before the occlusion. As the last step of the coast mode tracking, this paper presents a target re-locking algorithm to resume the target tracking after the blind time. First, during the normal image tracking stage, as a target model, a gray-level histogram ratio of the target and background is computed for each frame of the normal stage images thereby updating the target model at each time step. When entering the coast mode due to occlusion, we run the re-locking algorithm for each frame of the coast mode images so that it can immediately resume the tracking right after the end of the blind time. The re-locking algorithm divides the input image into blocks and for each block of the image, it takes an average of histogram ratios over the block and selects candidate target blocks of large histogram ratios, where the histogram ratio is evaluated at the gray-level of each pixel in the block and those histogram ratios are averaged over the pixels in the block. Due to the block-based averaging, the overall decision is robust to noise in the IR image, and the re-locking process afterward is of reduced computational complexity. With the target candidate blocks, a clustering is performed to make target candidate clusters, where each cluster is a set of connected blocks of large histogram ratios. As a final step, the first-ranked target candidate cluster is selected by computing an overall score that combines the histogram ratios and the prior knowledge of the target size, location, and variation of the intensity obtained during the normal tracking stage. We present experimental results based on both computer simulation and test under real environment with EOTS demonstrating the effectiveness of the proposed algorithm.
|