Loop Closure Detection for Visual SLAM Fusing Semantic Information

Abstract

Loop closure detection is of great significance to Visual Simultaneous Localization and Mapping (SLAM) system, which is used to correct accumulative errors in the process of robot motion. In this paper, the shortcomings and limitations of traditional loop closure detection methods in visual SLAM system are analyzed, and a loop closure detection method fusing semantic information is proposed. Faster R-CNN convolution neural network model for image target detection is applied to a traditional loop closure detection method to realize the fusion of semantic similarity and feature point similarity based on Bag-of-Words (BoW) model, and to judge loops by using the fused similarity. The method is tested on the open data sets. The experimental results show that the proposed method has better detection effect in dynamic scenes, can improve the accuracy and recall rate of loop closure detection, and the system has stronger robustness.