Pose Estimation Algorithm Based on Improved RANSAC with an RGB-D Camera

Pose estimation of moving body, such as mobile robots and UAVs(Unmanned Aerial Vehicles), is critical for visual Simultaneous Localization and Mapping(vSLAM). For the moving body carried with an RGB-D camera, their pose can be assessed through estimating the camera pose in RGB-D SLAM. Most RGB-D SLAM algorithms employ a framework with RANSAC and ICP(Iterative Closest Point) to refine feature correspondences and calculate camera pose simultaneously. However, due to its random sampling, standard RANSAC can not always find the optimal feature correspondences and performs not very well with lower accuracy and speed when false matches account for higher proportion. In this paper, we propose a new pose estimation algorithm based on RGB-D SLAM presented by Endres et al. [16] with ICP and improved RANSAC. Compared with the work of Endres et al., the improved RANSAC can be much more effectively to reject false feature matches, leading to more accurate pose estimation with faster speed. Several experiments are performed to prove the effectiveness of our algorithm.