두 유형의 경계 상자를 갖는 YOLO 네트워크 설계

초록

Object detection is a pivotal research area within computer vision, focusing on identifying and precisely localizing objects within images or video frames. Object detection algorithms like YOLO(You Only Look Once) traditionally employ rectangular bounding boxes to denote object locations, owing to their simplicity and computational efficiency. However, the variety of shapes and sizes of real-world objects often makes rectangular bounding boxes problematic. For circular objects, detection accuracy may be reduced due to the inclusion of unnecessary background information. This study proposes an enhanced YOLO network that incorporates both rectangular and circular bounding boxes to more accurately reflect the morphological characteristics of various objects. By selecting the optimal bounding box based on object shape, the network aims to improve detection efficiency and overcome the limitations of traditional systems. Therefore, the proposed network is expected to detect objects more accurately and efficiently and apply to multiple applications or systems.

키워드