재식별을 이용한 오토바이 추적 프레임워크 개발

초록

Tracking motorcycles in urban surveillance environments presents technical challenges due to their small size, high mobility, and frequent occlusions. In this paper, we propose a framework that enhances motorcycle ReID(Re-Identification) accuracy by applying algorithms robust to small object detection. First, real-time object detection is performed using YOLOv11, and to improve the representation and detection performance of small objects, we apply SF(Slicing Aided Fine-tuning) and SAHI(Slicing Aided Hyper Inference). The detected motorcycle objects are systematically grouped and annotated to build a robust ReID dataset. Using a transfer learning-based ReID model, each motorcycle is assigned a unique ID, and similarity-based identification is conducted. The proposed framework supports motorcycle tracking in multi-camera environments and query-based retrieval, enabling accurate identification through visual similarity ranking.

키워드