一种结合LATCH与多索引哈希的闭环检测方法

Loop Closure Detection Method Combining LATCH and Multi Index Hashing

摘要: RatSLAM是一个受海马体启发的同时定位与地图构建(SLAM)解决方案。针对RatSLAM由于一般的视觉处理方法而在复杂的室内环境中表现不佳的情况。本文首先提出了一种改进的RatSLAM，这是一种基于特征的单目SLAM系统，可以实时运行。其次，本文引入了多索引哈希(MIH)算法和三块区域代码(LATCH)的学习排列，用于评估当前场景与先前场景的相似性。最后，通过贝叶斯滤波和滑动窗口筛选出最相似的视觉模板，完成回环检测(LCD)。实验结果表明，改进后的系统在室内环境中的位置识别精度更高，体验地图的构建性能也更好，最大召回率达到60%。

Abstract: RatSLAM is a simultaneous localization and mapping (SLAM) solution inspired by the hippocampus. Regarding the poor performance of RatSLAM in complex indoor environments due to general visual processing methods. This article first proposes an improved RatSLAM, which is a feature-based monocular SLAM system that can run in real-time. Secondly, this article introduces the Multi Index Hash (MIH) algorithm and the Learning Arrangement of Three Block Region Code (LATCH) to evaluate the similarity between the current scene and the previous scene. Finally, the most similar visual template is selected through Bayesian filtering and sliding window filtering to complete loop closure detection (LCD). The experimental results show that the improved system has higher position recognition accuracy in indoor environments, better performance in constructing experiential maps, and a maximum recall rate of 60%.