2014 年 32 巻 9 号 p. 825-831
As of today, several rovers have been deployed on planetary surfaces and provided astonishing scientific achievements. A significant challenge in future missions is accurate localization for long distances to expand the activity range. After the successful Mars Exploration Rovers mission, visual odometry is regarded as a promising technique to provide accurate pose estimates on planetary surfaces. However, it should address the problem of terrain dependency since it cannot estimate motion in the terrain where visual features are lacked. This paper proposes a visual odometry system for untextured natural terrain which is devoid of features. Several key techniques are presented including a scheme for terrain-adaptive feature detection, and a motion estimation method using fewer feature points. The feasibility of the proposed techniques is supported by the field tests in a volcanic field using a test-bed rover.