Fusing Semantic Segmentation and Object Detection for Visual SLAM in Dynamic Scenes

Fusing Semantic Segmentation and Object Detection for Visual SLAM in Dynamic Scenes
Peilin Yu, Chi Guo, yang Liu, and Huyin Zhang

VRST 2021
Session: Paper 1: Tracking, Rendering and Social Interaction

Abstract
The assumption of static scenes limits the performance of traditional visual SLAM. Many existing solutions adopt deep learning methods or geometric constraints to cope with the dynamic objects in the scenes, but these schemes have problems of efficiency or robustness to a certain extent. In this paper, we propose a synergistic solution of object detection and semantic segmentation to obtain the prior contours of potential dynamic objects. With this prior information, strategies of geometric constraints are utilized to assist with removing dynamic feature points. Finally, the evaluation with the public datasets demonstrates that our proposed method can improve the accuracy of pose estimation and robustness of visual SLAM with no efficiency loss in high dynamic scenarios.

Videos for VRST 2021