Various hippocampal-entorhinal-based models are used to construct brain-inspired simultaneous localization and mapping (SLAM) systems. Loop closure detection (LCD) is a critical process of SLAM systems for robots to relocalize themselves… Click to show full abstract
Various hippocampal-entorhinal-based models are used to construct brain-inspired simultaneous localization and mapping (SLAM) systems. Loop closure detection (LCD) is a critical process of SLAM systems for robots to relocalize themselves and correct accumulative errors. The existing LCD methods of brain-inspired SLAM systems cannot solve well with challenging or large-scale environments by hand-crafted features and brute force search strategy. In this article, we propose a hybrid LCD method, which is based on the convolutional neural network (CNN) features and the fly’s locality-sensitive hashing algorithm (Fly-LSH). CNN features can improve the reliability of image matching results, and the Fly-LSH, a nearest neighbor search method derived from the fruit fly olfactory circuit, is used to accelerate the image processing. We use multiple hash tables to make a balance between the image matching time and the accuracy of loop closures. The proposed method provides a general approach for SLAM systems to process visual cues, and has application in a hippocampal-entorhinal-based SLAM system. The experimental results verify that the proposed method enables the system to build cognitive maps with better robustness and efficiency.
               
Click one of the above tabs to view related content.