Design and implementation of a RANSAC RGB-D mapping algorithm for multi-view point cloud registration

Abstract

This paper addresses multi-view point cloud registration, which is one of the major techniques to develop a robust visual tracking system. The proposed algorithm first uses Kinect camera to capture point clouds of an object-of-interest or a complex environment with different view points. Then, a novel RANSAC RGB-D mapping algorithm is developed to efficiently and accurately align the multiple point clouds. Based on the aligned point cloud of the object, a CAD model can possibly be created to achieve the purpose of three-dimensional object pose estimation, which helps improving the robustness of visual tracking systems.