Rover Localization through 3D Terrain Registration in Natural Environments

The registration of 3D points clouds is an important and challenging task in computer vision. In this paper we consider the problem of localizing a rover through 3D terrain registration in a natural environment. Two different local feature-based 3D terrain registration approaches are investigated: spin-image matching and point fingerprint matching. To overcome the huge memory storage problem of local features-based registration algorithms and improve the accuracy of the matching results, while reducing the computing time of the matching process, we developed an enhanced matching algorithm. The rover global localization scenario was conducted in the Mars Yard located at the Canadian Space Agency. The experimental results using natural environment data sensed by a high resolution and accurate 3D range sensor (LIDAR), demonstrate the effectiveness our enhanced matching algorithm