查看文章

This paper proposes a GPU (graphics processing unit)-based real-time RGB-D (red-green-blue depth) 3D SLAM (simultaneous localization and mapping) system. RGB-D data contain 2D image and per-pixel depth information. First, 6-DOF (degree-of-freedom) visual odometry is obtained through the 3D-RANSAC (three-dimensional random sample consensus) algorithm with image features. And a projective ICP (iterative closest point) algorithm gives an accurate odometry estimation result with depth information. For speed up extraction of features and ICP computation, GPU-based parallel computation is performed. After detecting loop closure, a graph-based SLAM algorithm optimizes trajectory of the sensor and 3D map.