A fast and accurate boundary tracking of moving objects in video

Extraction and tracking of moving objects in video is desired for many real-time applications such as motion capture and surveillance, while it remains a great challenge. In this paper, an efficient algorithm for boundary tracking of moving objects in video has been proposed, based on motion prediction and directional edge matching. Unlike traditional tracking methods based on clustering, the proposed algorithm can speed up the tracking processing, while maintain accurate as well. Experiment has been conducted to evaluate the performance of the proposed algorithm by comparing it with the well known mean shift algorithm, which is considered suitable for real-time tracking of non-rigid moving object. It is indicated from the experimental result that the proposed algorithm is both accurate and fast in tracking moving object, and robust to Gaussian noise to some extent, which could be qualified for realtime applications.