A novel hybrid visual odometry using an RGB-D camera

Abstract

RGB-D cameras provide both a color image and a depth image which contains the real depth information about per-pixel. The richness of their data and the development of low-cost sensors have combined to present an attractive opportunity for mobile robotics research. In this paper, we describe a novel hybrid visual odometry using an RGB-D camera. Different from the original method, it is a pure visual odometry method without any other information, such as inertial data. The important key is hybrid, which means that the odometry can be executed in two different processes depending on the conditions. It consists of two parts, including a feature-based visual odometry and a direct visual odometry. Details about the algorithm are discussed in the paper. Especially, the switch conditions are described in detail. Beside, we evaluate the continuity and robustness for the system on public dataset. The experiments demonstrate that our system has more stable continuity and better robustness.