Abstract
Estimating a robot's own position is an essential requirement for autonomous mobile robots. Visual Odometry (VO) can play a key role in navigation tasks in slippery natural terrain, which highly degrade the accuracy of Wheel Odometry, without relying on other infrastructures nor any prior knowledge. A challenge of VO system lies in how to increase the number of features tracked between continuous frames in untextured terrain. Although there has been many feature selection algorithms proposed, none of them are sufficient in terms of calculation speed and robustness. Hence, this paper propose an algorithm that dynamically switches between high-speed/low-accuracy and low-speed/high-accuracy algorithm according to the texture of terrain. Validity of this algorithm was proved by the simulation using datasets taken at Izu-Oshima, Japan.
