Visual Localization Using 3D Gaussian Splatting Representation for Mobile Robots With Geometric Feature Correspondences Synthesis

Achieving visual localization with excellent interactive performance is challenging for mobile robots. Based on real-time photo-realistic view synthesis, 3D Gaussian splatting (3DGS) representation has demonstrated vast potential for robots engaging with the physical world. In this work, we propose a novel coarse-to-fine visual localization method named L3DGS based on the 3DGS radiance field representation. Particularly, during the coarse stage, we exploit novel views synthesized by the pretrained 3DGS map to create geometric feature correspondences to perform geometric alignment. Then, we integrate both geometric and photometric alignment to refine the camera pose. Unlike previous radiance field-based approaches, we leverage geometric feature correspondences and the innovative 3DGS map to improve the localization accuracy. In our experiments, we evaluate the proposed method across two real-world indoor and outdoor datasets. Consequently, compared to the baselines, the proposed method achieves competitive or superior experimental results. Note to Practitioners—This paper is motivated by the problem of map-based visual localization for autonomous mobile robots but is also applied to other robotic systems that represent a 3D scene using 3D Gaussian splatting (3DGS) representation. In practical applications, the accuracy of existing radiance field-based localization methods is susceptible to image retrieval. In this paper, we propose a visual localization method applied to both indoor and outdoor scenarios for mobile robots. Our technical innovation lies in using 3DGS as the map and utilizing its rendered images to enhance the robustness of the visual localization algorithm. Compared with structure-based methods, the proposed method requires less time for 3D map reconstruction, while enhancing both the accuracy and robustness of visual localization. We conduct extensive experimental studies in real-world indoor and outdoor scenarios to validate the effectiveness of the proposed method. In the future, we will focus on improving the efficiency of the proposed method and carry out applications in visual navigation and path planning of mobile robots.