An RGBD-SLAM with Bi-directional PnP Method and Fuzzy Frame Detection Module

The core technology to mobile robots is referred to as Simultaneous Localization and Mapping (SLAM). It has been wildly investigated on improving pose estimate accuracy and mapping consistency. This paper proposes several algorithms on the target. Firstly, a fuzzy frame evaluation and filtering strategy based on OpenCV is proposed to prevent fuzzy frames from affecting pose estimation. Secondly, a Bi-directional PnP-RANSAC method is proposed to estimate the pose between frames, in which more matching information is exploited, and the consistency of results can be verified to obtain more accurate pose results. Furthermore, for the consideration of efficiency, a GPU acceleration algorithm named SiftGPU [1] is applied in the feature extraction and matching process to improve real-time performance. Finally, a complete RGB-D SLAM framework is built to realize the above algorithms. Experiments have been conducted in the lab environment and dataset [2], producing global consistent dense 3D environment representation and accurately estimated trajectory, demonstrating the effectiveness of the proposed methods.