A real time optimization-based SLAM for indoor UAV flying robots

Vision-based simultaneous localization and mapping in GPS-denied and indoor environments is one of the most important requirements for robot navigation. One of the major challenges to real-time implementation is the limited processing power on small flying robots. In this article, a method for enhancing RGB frames using histogram equalization and blind/reference less image spatial quality evaluator is presented. Some publicly available RGB-D dataset are employed to evaluate the proposed technique with our hybrid method using real-time appearance-based mapping. The problem is expressed as a graph configuration and is optimized using the Levenberg-Marquardt optimization approach. The results of evaluation on online data sets and experiments using designed flying robot indicate reduction in absolute translational error in comparison with the ground truth. Also, the accuracy and robustness of the proposed method have improved at least 17% with respect to environmental conditions and system noise.