Structural Features based Visual Odometry for Indoor Textureless Environments

Abstract

Simultaneous localization and mapping (SLAM) is one of key technologies for autonomous mobile robots. The classical feature-based SLAM methods mostly use point features or line features. In the indoor artificial environment with low texture information, these features are often insufficient, which leads to the method's failure. In this paper, the vertical and coplanar constraints between the structural features of points, lines, and planes are parameterized and integrated with the traditional point-based method to estimate the pose. Public benchmark datasets are employed to evaluate the advanced method together with the state-of-the-art methods. The experimental results show that better accuracy can be achieved by the advanced method compared to the state-of-art in low texture environment.