Open Access
Research on Pedestrian Multi Object Tracking Based on FairMot Transfer Learning
Author(s) -
Hongjun Chen,
Liang Zhao,
Lei Ma
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1846/1/012030
Subject(s) - object (grammar) , tracking (education) , artificial intelligence , computer vision , computer science , video tracking , transfer of learning , object detection , transfer (computing) , process (computing) , viola–jones object detection framework , pattern recognition (psychology) , psychology , pedagogy , parallel computing , face detection , facial recognition system , operating system
Object tracking is divided into two processes: multi object detection and multi object tracking. In this paper, the multi object detection is studied firstly, and the multi object detection algorithms based on prior box anchor and anchor-free are analyzed and compared. Then in the second process of multi object tracking, multi object tracking algorithm are researched: the multi object tracking algorithm based on prior box anchor, the multi object tracking algorithm based on anchor-free. Finally, Finally, an experiment is carried out to track the pedestrians on the road by using FairMot transfer learning, and the metrics of MOTA, IDF1, IDS, MT, ML and FPS are analyzed. Experiments show that under the method of using transfer learning, MOTA can be improved by 4.2 percentage points.