Machine Learning report - Project II
88 Citations•2021•
Bartlomiej Binda, Clément Petit, Yanis Berkani
journal unavailable
It is concluded that the best compromise was a model using SSD detection and MOSSE tracking : the authors managed to achieve 31 fps and 63% of accuracy.
Abstract
—This report focuses on multi-object detection in image sequences. Throughout the paper, several detection algorithms (namely YOLO and SSD) and tracking algorithms (namely MOSSE, KCF and Deep SORT) are compared. Furthermore, the limitations of the algorithms are discussed, along with measures of accuracy and speed. We finally concluded that the best compromise was a model using SSD detection and MOSSE tracking : we managed to achieve 31 fps and 63% of accuracy.