Robust Loop-Closure Algorithm Based on GNSS Raw Data for Large-Scale and Complex Environments

Trajectory drift is a significant issue in intelligent vehicles that rely on multi-source fusion for achieving autonomous navigation, hindering the accurate positioning of these vehicles. Existing loop-closure algorithms mostly rely on single-sensor data, such as visual or LiDAR (Light Detection and Ranging) data, making their performance susceptible to degradation in challenging environments. This paper proposes a novel loop-closure detection algorithm. It employs ORB (Oriented FAST and Rotated BRIEF) feature points from visual data for loop-closure candidate frame detection, confirms loop-closure frames using depth values mapped from LiDAR data onto images, and finally utilizes GNSS (Global Navigation Satellite System) raw data to validate and correct trajectory drift further. The paper concludes by validating the stability and robustness of the proposed loop-closure algorithm through simulations with various publicly available datasets, thereby assessing its performance in large-scale and complex environments. Furthermore, by integrating the proposed loop-closure algorithm into an openly accessible online odometer, the results demonstrate an improvement in position estimation accuracy by 83.45%.