Dynamic Adaptive Parametric Multi-Target Tracking Algorithm Based on Appearance and Pose Feature
作者
Jiale Hu,Wenli Zhou,Fang Liu,Gang He,Ruoning Song
出处
期刊:日期:2023-11-03卷期号:: 107-111
标识
DOI:10.1109/ic-nidc59918.2023.10390705
摘要
In recent years, multi-target tracking missions have evolved considerably. Nevertheless, the fusion of pedestrian appearance and pose feature into association models is an area that has received insufficient attention. Furthermore, despite efforts to address ID switches through techniques like SORT, their effectiveness has been somewhat constrained. In this paper, a distinctive method is introduced wherein pedestrian appearance and pose features are combined in the association algorithm, complemented by the utilization of dynamic adaptive parameters to further optimize its performance. Through extensive experimentation on widely acknowledged datasets, including MOT17, MOT20, and DanceTrack, our approach demonstrates remarkable achievements. Our approach effectively decreases the occurrence of ID switches by approximately 50%. We are confident that our study sets a new standard and offers a novel viewpoint for the domain of multi-target tracking. The GitHub repository provides access to and allows for the downloading of the source code: https://github.com/WeiXiao-Hyy/shuangmu_renyuan_model.