A static feature points detection algorithm for visual odometry using optical flow vectors

This paper proposes a robust and precise method for detection the static feature points in dynamic scenes. We use the optical flow vectors as the basis for distinguishing the static features from the moving ones. This method is suitable for the visual odometry systems of vehicles and robots. The static point is selected by a background point set determination process including motion clustering and motion recognition. The motion clustering separates moving objects from background according to optical flow orientation, and the motion recognition determines a background cluster according to scatteredness in image coordinates. The approach presented here is tested on substantial videos and the results prove the robustness and precision of the method.