Efficient and Accurate Two-Step Sequence-Based Loop Closure Detection

Loop closure detection is a critical issue in Visual Simultaneous Localization and Mapping (VSLAM), as it involves determining when a robot revisits previously explored locations. To improve the accuracy and real-time performance of loop closure detection, this paper proposes a two-step loop closure detection method based on keyframe sequences. It dynamically partitions the keyframes into sequences based on the angular difference between adjacent keyframes. Sequence-to-sequence matching is performed using the Bag-of-Words (BoW) model. To reduce the time overhead of loop closure detection, we use histograms during the frame-to-frame matching process. This allows us to avoid spending too much time on the extraction and matching of complex descriptors in matched keyframe sequences, while ensuring the accuracy of the results. To further enhance efficiency, we design a filtering mechanism based on past loop closure results and sequence length, which avoids unnecessary detection in frames where loop closures are unlikely to occur. We integrate the proposed loop closure detection approach into VINS-Mono and evaluate it using public datasets. Experimental results demonstrate that our loop closure detection approach outperforms the original approach in terms of both accuracy and computational efficiency, making it well suited for real-time large-scale VSLAM applications.