3-D Mapping and Localization using RGB-D Camera

This paper presents a real-time 3-D mapping system based on RGB-D camera. Our system concurrently estimates the trajectory of RGB-D camera and generate a dense 3D model of the surrounding. From RGB and Depth video sequence, our propose method uses 2D-2D feature base visual odometry to computes the 6-DOF camera pose utilizing k-nearest neighbor feature matching. After obtaining pairwise features, we propose infinity point filter to remove infinity pair points, which leads motion estimation produce a trivial solution. Then in the 2D-2D motion estimation procedure, we present the scene model selection algorithm to deal with frame scene that be planar or almost planar, which makes the Essential assumption gives the wrong answer. The experiments result shows a successfully built 3D map without non-linear optimization and loop close detector for real-world environment.