가중치 기법을 적용한 Mask R-CNN 기반 객체 검출 시스템 구현

Abstract

The fields of computer vision and image processing are rapidly evolving, fueled by improvements in hardware performance and advancements in deep learning algorithms. Object tracking, in particular, is gaining traction, playing a key role in diverse applications such as video surveillance, autonomous driving, and computer vision. Existing object tracking techniques are primarily based on background subtraction or frame differencing, but they suffer from significant degradation in tracking accuracy in complex environments, such as lighting changes, subtle background movements, and the reappearance of stationary objects. To address these issues, this paper applies the Mask R-CNN algorithm to a weighting technique that enables precise object tracking and recognition even in constantly changing backgrounds. This approach demonstrates its potential to enhance the reliability of image processing systems and improve object tracking performance. Furthermore, we propose an experimental verification method that effectively distinguishes between object and background changes in a variety of environments.