Improved-ORB-based SLAM Using RGB-D Data

RGB-D cameras have become a popular sensor for Simultaneous Localization and Mapping (SLAM) because they can provide RGB and depth information at the same time. The spatial location of feature points of SLAM is obtained by its corresponding depth information from RGB-D sensors. However, the depth information of the feature at the edge of the object is very noisy, and the measurement is often inaccurate. At the same time, there is poor real-time performance, low accuracy, and much difficulties of dense construction in ORB-based Visual SLAM (visual SLAM, VSLAM). Therefore, an improved ORB features based SLAM framework with corresponding method using RGB-D data is proposed in this paper. We extract and match features with improved ORB. Then the local map is used to perform local optimization for pose estimation, and an efficient key frame screening method is designed. The proposed method effectively reduces errors, enhances robustness, and builds a dense map of the environment.