차량 내 다중 센서 시간 오차 정합을 위한 시멘틱 기반 오프셋 추정 방법

Abstract

Accurate sensor synchronization is critical for multi-sensor fusion in autonomous vehicles. Time offsets between sensors such as IMU, LiDAR, camera, and RTK-GNSS can significantly degrade localization and tracking performance. This paper presents an Extended Kalman Filter-based method for estimating and correcting inter-sensor time offsets in real-time. Each sensor’s odometry is fused into a common frame, and time delays are modeled as part of the EKF state vector. IMU data is used for prediction, while visual odometry (VO), LiDAR odometry (LO), and RTK measurements serve as delayed observations. To enhance adaptability across varying environments, we incorporate a semantic modeling framework based on the Triplet Ontological Semantic Model (TOSM). Each sensor is described by its implicit, explicit, and symbolic properties, and is linked to place-specific context through a semantic HD map. This enables context-aware fusion strategies that consider both sensor characteristics and environmental semantics. We validate the proposed method using real-world driving data collected from a vehicle equipped with multiple sensors. Experimental results show improved alignment and reduced fusion errors after time offset correction. The framework is lightweight, runs efficiently on embedded systems, and supports both online and offline calibration. Overall, the method improves robustness and synchronization accuracy in real-world autonomous driving scenarios.