A Brain-inspired SLAM System Based on Keyframe Template Matching

NeuroSLAM is a novel brain-inspired Simultaneous Localization and Mapping (SLAM) system. The traditional NeuroSLAM algorithm is significantly influenced by environmental factors and lighting conditions, resulting in decreased mapping accuracy and stability. Therefore, a novel approach based on keyframe template matching has been proposed to overcome the deficiencies in visual processing within the original NeuroSLAM system. We initially introduce ORB features as a means to select keyframes, aiming to mitigate issues of redundancy and loss of keyframe information caused by fixed step lengths. Subsequently, we combine the bag-of-words model with a visual template matching algorithm to construct a keyframe template database, enabling us to achieve the objective of online, accurate, and rapid environment recognition. The experimental results indicate that compared to NeuroSLAM, our method significantly reduces loop closure detection time and improves loop closure recall rate. Additionally, in comparison to the ORB-SLAM3 algorithm, our method demonstrates greater robustness.