A method of estimating motion trajectory with combining particle filter and optical flow

In our previous research, it was proved that accuracy and efficiency of object tracking increase when we combined a particle filter and optical flow. At a boundary between object and background in the image, however, trajectory of motion turned away from the correct direction. By using Hesse matrix that consists of the derivatives about the brightness, each particle´s position was decided whether it was near the boundary or not, and it enabled particles to avoid such area. As a result, the incorrect trajectories of motion were decreased. In addition, by improving the resampling process, particles have correct velocity were increased. However, in a particular object, its trajectories were broken off on the way. In brief, the tracking failed in regard to such an object. So far, the regions whose exact speed was easy to calculate could be tracked appropriately, but the objects whose inside consist of parts such as edges could not be tracked and their trajectories were broken off.