SV-Loop:Semantic-Visual Loop Closure Detection with Panoptic Segmentation

Loop closure detection plays an important role in simultaneous localization and mapping (SLAM) system. Traditional visual-based methods of loop closure detection are hard to relieve the issue of dynamic objects and change of light in complex environments. To improve robustness of loop closure detection, a loop closure detection algorithm named SV-Loop combines semantic and visual information is proposed in this paper. This work leverages the latest panoptic segmentation network to extract semantic information. Firstly, similarity of details is calculated through visual module in micro perspective. Then similarity of objects is computed by semantic module from a macro perspective. Finally, visual and semantic loop candidates are integrated to determine loop closure. Experiments have been conducted on open source datasets and practical scene. Experimental results have shown higher accuracy and better robustness of proposed method in complex environmental changes.