Research on Building Density Map Construction with Improved ORB-SLAM2

ORB-SLAM2 (Oriented FAST and Rotated BRIEF SLAM) is an efficient visual SLAM algorithm widely used in fields such as robot navigation and autonomous driving. However, when constructing dense maps in complex architectural environments, ORB-SLAM2 faces challenges related to feature point extraction and mapping accuracy. This paper proposes an improved ORB-SLAM2 system based on RGB-D cameras, optimizing feature point extraction with middleThFAST thresholding and incorporating the YOLOv4-tiny object detection algorithm to enhance map accuracy and stability. Experiments validated on the Mikolajczy public dataset show that the improved ORB-SLAM2 system achieved a 15.5% improvement in feature point extraction time efficiency, approximately 10% better uniformity in feature point distribution, and an 86.42% reduction in trajectory error in dynamic scenes compared to the traditional ORB-SLAM2, demonstrating high robustness in highly dynamic environments. This research provides a more efficient solution for constructing three-dimensional dense maps in complex architectural scenarios for autonomous navigation.