Development of Visual Odometory Method Based on Robust Tracking of Multiple Feature Points on Floor Surface in Intensity Change(Mechanical Systems)
Masashi Yokozuka, Koichi OZAKI · TRANSACTIONS OF THE JAPAN SOCIETY OF MECHANICAL ENGINEERS Series C · 2010
This paper proposes a novel visual odometry method by tracking feature points on floor surfaces. In floors, there are many shadows by some objects and image changes in tracking region when a view point moves. It is necessary for robust tracking to extract stably feature points on the floor in intensity change. The proposal method extracts the feature points by using various differential filters which output constant values in intensity change. We normalize the filters in intensity change for robust tracking. Generally, template matching is used in tracking methods for visual odometry. Measuring resolution and limitation of the methods depends on the number of templates. Robots' motions are constrained by the resolution and limitation constrain. In order to reduce the constraint, the proposal method measures motion by using the least-square method of estimation motion errors from a large number of tracking points on motions. The experimental results show that the proposal method by using normalized filters can track stably the feature points in intensity change and can measure a round path repeatedly by practicable accuracy.