Temporal-Aware Hybrid Energy Optimization: Surface Loss and Differentiable SDF in SLAM

In 3D scene reconstruction and SLAM (Simultaneous Localization and Mapping) systems, precise camera pose estimation and high-quality scene rendering are critical. Traditional approaches often overlook the temporal continuity of camera motion, potentially compromising reconstruction quality and the accuracy of pose estimation. This paper introduces a dual-energy function optimization method that integrates surface loss with differentiable Signed Distance Function (SDF) rendering. This approach enhances pose estimation and scene reconstruction quality by considering temporal continuity. Initially, we propose a surface loss function based on temporal continuity, designed to optimize inter-frame continuity during camera pose estimation. Subsequently, we develop a novel temporally continuous differentiable SDF renderer, aimed at more effectively handling dynamic changes in the scene and detail reconstruction. These two energy functions work in tandem to provide more accurate and coherent pose estimation and scene reconstruction. Experiments conducted on various challenging datasets demonstrate that our method surpasses existing techniques in reconstruction precision, detail preservation, and computational efficiency. Notably, our approach exhibits greater robustness and accuracy, especially in scenarios involving rapid movement and occlusions.