Improvement of Accuracy and Robustness on Stereo Imaging Method by Using Shadow

Abstract

Robotics has attracted much attention from the planetary surface exploration community ever since NASA 's Pathfinder Mission opened a new door for the subject. In order to collect geological samples, 3D surface information of the environment and the targets, in general, has to be obtained so that a manipulator can successfully grasp them. Measurement accuracy and robustness are key issues for the realization of autonomous sample acquisition. A stereo camera is generally a useful device for 3D surface measurement; however, it may not be robust enough to handle unknown natural objects that exhibit a great deal of similarity and where there is uncertainty over their texture. In this paper, we focus on the robustness and accuracy of a measurement method using a stereo camera, and we propose Shadow Casting Stereo Imaging (SCSI). In use, SCSI casts the shadow of the manipulator (or a stick-shaped object) on a target surface, and then the 3D information can be measured using conventional stereo imaging by utilizing this cast shadow. The shadow creates a featured area on similar textures under harsh lighting conditions, because it produces a darker area on the textures as feature points for image processing. In addition, this technique reduces the search area for the template matching, and lowering calculation costs. We evaluated the performance of SCSI by comparing the measured results with conventional stereo imaging. Our results show that SCSI has advantages in terms of accuracy, robustness, and calculation cost against similarity of a natural object under various lighting conditions.