複数台監視カメラによる人物トラッキングシステムの開発

抄録

Various methods have been proposed for person tracking by image recognition. This paper aims to improve the accuracy of person tracking with image features using deep learning models. In person tracking using image features based on deep learning models, identity is determined by focusing on the color of the image. However, the problem was that noise, such as overlapping between persons or unclear images with few pixels, reduced the accuracy of judgment. Therefore, the proposed system performs robust re-identification using two evaluation metrics, one based on image features and the other on Intersection over Union.