Pedestrian Multiple-object Tracking Based On Fairmot And Circle Loss

This paper proposes a method for multi-object tracking using the High-Resolution Net (HRNet) as the baseline. The HRNet is used to learn high-resolution representations with strong position sensitivity and is capable of processing multiple resolution network branches in parallel and continuously carrying out information interaction between different branches. The proposed method is shown to improve accuracy and robustness compared to existing advanced algorithms on the MOT17 dataset.