차량 위치 추정을 위한 CNN-LSTM 기반 추측 항법 오차 보상 모델

다국어 초록 (Multilingual Abstract)

Vehicle localization becomes challenging in GNSS-denied environments due to the accumulation of dead-reckoning errors. A learning-based yaw-error compensation method using only vehicle speed and yaw-rate signals is presented. Temporal features are extracted by a CNN-LSTM, and multiple experts generate candidate yaw-error estimates. These estimates are combined through deterministic interpolation in the speed and yaw-rate state space to reflect state-dependent error characteristics. The predicted yaw error is used to correct the dead-reckoning heading and reconstruct the vehicle trajectory during GNSS outages. Experiments on low-speed driving data confirm that the proposed method reduces accumulated localization error and improves trajectory estimation performance.

목차 (Table of Contents)

• Abstract
• 1. 서론
• 2. 차량 모델 및 학습 데이터 구성
• 3. 제안 방법
• 4. 실험
• 5. 결론
• References