Lidar 지도를 이용한 시각적 위치 추정 방법론 비교

Abstract

This paper explores visual inertial odometry (VIO) using a Lidar prior map, focusing on the comparison between loosely coupled and tightly coupled approaches. The study investigates how the integration of visual and inertial data, combined with Lidar information, enhances localization performance, particularly under varying conditions of feature matching reliability. A comprehensive preprocessing pipeline is proposed, including reference image alignment and depth map generation, to support accurate visual relocalization. The integration of stereo cameras, IMU data, and wheel odometry is examined within both VIO frameworks. The methods are tested using a complex urban dataset to evaluate their effectiveness in real-world autonomous driving scenarios. By comparing the performance of each approach, this research provides insights into their applicability in choosing method. The performance of the two approaches is assessed under different conditions to determine which provides the most reliable results for VIO tasks.