차량 정보 인식 및 추적 기법을 이용한 인공지능 학습용 교통정보 데이터 자동정제

초록

Artificial intelligence services are currently becoming increasingly integrated into our lives. Hence, the collection of quality data is becoming crucial. In particular, artificial intelligence learning models that use traffic data can be employed in various fields, including navigation, traffic information, autonomous driving, traffic accident prevention, and criminal vehicle tracking. Hence, collecting high-quality learning data and developing efficient refinement methods to improve performance are essential. Therefore, this paper proposes an automatic refinement method of traffic information data for artificial intelligence learning. We used YOLOv5 object detection, IOU tracking, vehicle classification, and variable thresholds in a parallel algorithm for license plate recognition to perform data refinement and demonstrated savings in terms of both time and cost, compared with manual work. The results confirm that various refinement methods of vehicle information recognition and tracking methods are more effective for quality data refinement.

키워드