Robust visual tracking using a spatio-temporal approach with optical flow

Most visual tracking algorithms need more or less features of moving objects such as shape, color, dimension, and speed, etc. A three-dimensional filter based on network resonance has been developed for detection of moving objects with a specific moving velocity. This spatio-temporal approach is able to accomplish tracking tasks for unknown moving objects with overlapped or crossover path. However, constraints of smooth moving trajectory and back ground removal confine its applications. Therefore, this paper incorporates the optical flow technique into the spatio-temporal scheme to overcome those difficulties. Experiments on synthesized images and natural images are conducted to demonstrate improved tracking performance using the proposed method.