A Modular Loop Closure Detection Scheme for Autonomous Driving: A Loosely Coupled Approach

Efficient and precise loop closure detection is essential for both autonomous vehicles and robotics. Currently, loop closure detection technologies can recognize locations using the similarity between environmental measurements. However, the inherent errors in these measurements present a significant challenge to detection performance, which has not been adequately addressed in the previous literature. To address these issues, this paper proposes a novel modular loop closure detection scheme based on the temporal similarity of loop sequences to enhance the consistency of constructed maps and thereby improve detection performance. Specifically, the similarity and credibility of loop closure sequences, as well as their corresponding criteria, are defined based on the characteristics of loop closure occurrences in map construction. Fast filtering and accurate matching strategies have been developed based on the similarity and credibility of loop closure sequences. An independent recall strategy is then developed to mitigate the impact of measurement similarity errors on recall. Unlike traditional approaches, this method integrates temporal similarity with existing measurement similarity schemes in a loosely coupled manner, improving the performance and efficiency of loop closure detection in complex environments. The effectiveness of the proposed method is sufficiently validated using both the public KITTI dataset and real field-test vehicles. The results demonstrate significant improvements in loop closure detection, providing a solid foundation for the advancement of autonomous driving technologies.