| Target tracking technology is an important technology in the field of computer vision, which has been widely used in various fields. The traditional target tracking algorithms generally specific for a particular scene or track targets can not be achieved for any target tracking, what’s more, the traditional tracking algorithm in tracking process will lead to tracking failure accumulation of errors which can not be a long track. Meanwhile, with the rapid development of computer technology, target tracking algorithm gradually be applied to embedded systems, target tracking algorithms in embedded systems to meet real-time requirements and maintain a long track steadily become a hot topic in the field of computer vision problems.In this paper, a large number of target tracking algorithm are researched and we find tracking-learning-detection algorithm has good robustness. on this basis of this algorithm, we propose a parallel algorithm to enable real-time processing of HD video image and port the algorithm to embedded systems, which can be more widely used. In this paper, the specific work location is as follows:1. On the tracking-learning-detection algorithm in-depth study, a detailed analysis of the median flow tracking algorithm,the detection algorithm based on random forest and the learning mechanism based on P-N constrain, moreover, a large number of experiments are made to show this algorithm stable and real-time.2. TLD strategy is insufficient to meet the real-time of high-definition video image processing, we use multi-process MPI and multi-thread OpenMP to optimize the TLD so that it can process the image data in parallel, which can facilitate more efficient use of the platform computing resources. We perform quantitative analysis by experiment the algorithm3. The TLD algorithms will be ported to more generic embedded platform, so we need to install the system and some software relied on an embedded platform, finally we verify its tracking by some experiments. |
PDF Full Text Request