ECH: An Enhanced Cart Histogram for place recognition

Place recognition is a crucial technique for mitigating accumulated errors in simultaneous localization and mapping (SLAM). However, this remains an immensely challenging task. Existing place recognition algorithms are susceptible to failure during translation or rotation changes, lacking both translation and rotation invariance simultaneously. In this paper, we adopt and enhance the global descriptor by incorporating principal component analysis (PCA) to align the current keyframe with the query keyframe to a consistent azimuth angle. Furthermore, to ensure algorithm speed, our method is primarily based on non-ground points, and we propose a novel iterative-based fast non-ground point extraction method. We extensively compare our proposed algorithm with other state-of-the-art algorithms (Scan Context, Intensity Scan Context, STD). All experimental results demonstrate that our proposed algorithm achieves higher precision and recall rates in loop closure detection.