Adaptive Approach to Simultaneous Localization and 3D Mapping Using an RGB-D Camera

An adaptive system for 3D environmental mapping using an RGB-D camera, designed to improve real-time visual odometry, is proposed. The system has two main innovations: an adaptive scheme for selecting a 2D feature detector and an adaptive process for specifying the camera position. The system dynamically selects the most suitable detector based on the environment conditions, increasing the number of processed key points even in scenes with limited visual characteristics. This leads to a more accurate 3D reconstruction of the surrounding space. In addition, adaptive specification of camera position optimizes image alignment and minimizes trajectory errors. Simulation results on the well-known TUM database show that the proposed system processes complex scenes with loop closures and different illumination conditions in real time.