2019/10/13 by Wei Wang, Wang, Wei, Muhamad Risqi U. Saputra +15 · 2 citations
Computer Science · Earth and Planetary Sciences · Engineering · #3D Surveying and Cultural Heritage #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics (cs.RO) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1910.11088
openalex publication_date 2019/10/13 · openalex created_date 2020/07/23 · openalex updated_date 2026/07/28
Odometry is of key importance for localization in the absence of a map. There\nis considerable work in the area of visual odometry (VO), and recent advances\nin deep learning have brought novel approaches to VO, which directly learn\nsalient features from raw images. These learning-based approaches have led to\nmore accurate and robust VO systems. However, they have not been well applied\nto point cloud data yet. In this work, we investigate how to exploit deep\nlearning to estimate point cloud odometry (PCO), which may serve as a critical\ncomponent in point cloud-based downstream tasks or learning-based systems.\nSpecifically, we propose a novel end-to-end deep parallel neural network called\nDeepPCO, which can estimate the 6-DOF poses using consecutive point clouds. It\nconsists of two parallel sub-networks to estimate 3-D translation and\norientation respectively rather than a single neural network. We validate our\napproach on KITTI Visual Odometry/SLAM benchmark dataset with different\nbaselines. Experiments demonstrate that the proposed approach achieves good\nperformance in terms of pose accuracy.\n