6DoF-SLAM using 3D Point Cloud-based Objects Recognition

Abstract

A method for three-dimensional (3D) point cloud-based object recognition and a method that uses the recognized objects for six-degree-of-freedom simultaneous localization and mapping (SLAM) with a high accuracy are presented. For object recognition, we use a convolutional neural network to identify the meaning of each point inside an input 3D point cloud. For scan registration, we present a highly accurate hybrid method that combines the iterative closest point with particle swarm optimization (PSO) to match the recognized points to be archived. Using PSO to match the recognized object's points in each neighboring scan can help decrease incorrect correspondences and enhance the robustness of scan matching. Compared to state-of-art methods, the proposed method achieved good performance on the KITTI odometry benchmark and our SLAM experiments.