Improvements in visual odometry algorithm for planetary exploration rovers

Localization is a key aspect of planetary exploration rovers. Visual odometry algorithm can give improved motion estimates compared to conventional systems (Wheel encoders/Inertial measurement units) when traveling in Steep/Sandy terrains. This paper gives the details of a Visual odometry algorithm developed for future planetary exploration rovers. A new algorithm to improve the feature matching and a computationally efficient method to compute the Jacobian used to model the triangulation errors are given here. The test results indicate good accuracy in estimation for the translational movement.