Fast Visual Odometry Based Sparse Geometric Constraint for RGB-D Camera

Ruibin GUO  Dongxiang ZHOU  Keju PENG  Yunhui LIU  

IEICE TRANSACTIONS on Information and Systems   Vol.E102-D   No.1   pp.214-218
Publication Date: 2019/01/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.2018EDL8119
Type of Manuscript: LETTER
Category: Image Recognition, Computer Vision
pose estimation,  fast visual odometry,  geometric cost function,  iterative optimization,  3D reconstruction,  

Full Text: FreePDF(1.3MB)

Pose estimation is a basic requirement for the autonomous behavior of robots. In this article we present a robust and fast visual odometry method to obtain camera poses by using RGB-D images. We first propose a motion estimation method based on sparse geometric constraint and derive the analytic Jacobian of the geometric cost function to improve the convergence performance, then we use our motion estimation method to replace the tracking thread in ORB-SLAM for improving its runtime performance. Experimental results show that our method is twice faster than ORB-SLAM while keeping the similar accuracy.