Heterogeneous SLAM Fusion Method in Complex Illumination Scenes

Aiming at the problems of closed-loop detection failure and low robot trajectory accuracy in simultaneous localization and mapping (SLAM) in complex lighting environments such as low light and weak texture, a heterogeneous SLAM fusion method based on fuzzy neural network (FNN) is proposed by combining the high-precision mapping and precise positioning ability of traditional visual SLAM methods with the strong scene recognition ability of bionic SLAM methods in complex illumination environments, including the decision method based on standard FNN which improves the success rate of loop-closure detection, and the trajectory optimization method based on T-S (Takagi-Sugeno) FNN which improves the accuracy of robot trajectory estimation, so as to achieve more accurate positioning and more reliable environment modeling results. Experimental results show that compared with ORB-SLAM2 and RatSLAM methods, the proposed heterogeneous SLAM fusion method has higher recall rate of loop-closure detection and lower absolute trajectory error (ATE) on selfcollected and public datasets, shows strong robustness in complex scenarios, and is promising in promoting the accuracy of autonomous robotic operation and the robustness of navigation and localization in complex illumination scenes.