Online indoor visual odometry with semantic assistance under implicit epipolar constraints

Yang Chen, Lin Zhang, Shengjie Zhao, Yicong Zhou

Abstract: Highlights•The first online semantics-aided VO for indoor environments, SEOVO, is proposed.•Semantic assistance is introduced to help adaptively deal with the challenging textureless regions.•A novel epipolar constrained photometric loss is designed to guide the pose regression.•Fusion of multiple cues makes our SEOVO outperform the state-of-the-art self-supervised VOs.