Tracking and Mapping Strategy for Indoor UAV Based on Entropy Theory: An ORB-SLAM3 Extension

For indoor application, which is an entirely GPS-denied environment, visual simultaneous localization and mapping (SLAM) facilitates the real autonomy of unmanned aerial vehicle (UAV) but raises the challenging requirement on accurate localization with low computational cost. To address this difficulty, a stereo-camera based SLAM system is proposed by applying Entropy theory to ORB-SLAM3. As an extension to ORB-SLAM3, the additional entropy decision module and map processor are specifically designed. The decision module can improve computing efficiency by deciding whether keyframes or extra optimization should be introduced. Furthermore, the map processor is targeted at loading and maintaining the prior map whenever needed. Experiment results in indoor laboratory environment show that the developed system can achieve the superior localization accuracy in more efficient computation manner with smaller size of mapping compared with ORB-SLAM3. Furthermore, the map can be effectively expanded and corrected even when prior information is invalid, greatly increasing the robustness of SLAM system.