GeekAlexis / FastMOT

High-performance multiple object tracking based on YOLO, Deep SORT, and KLT 🚀
MIT License
1.12k stars 256 forks source link

How to share lost track buffers across cameras for ReID? #250

Closed AngelaYZhang closed 1 year ago

AngelaYZhang commented 2 years ago

I've looked into #129 and #157 and managed to run two instances on jetson nano independently using multiprocessing. I would like the person to be re-identified again across cameras (the cameras have non-overlapping view, so they will disappear from one cam and appear in another). Currently I've only figured out self.hist_tracks is needed, what other lost track buffers should I share?

Also what is a good approach for sharing the variables with self.hist_tracks an OrderedDict. I've tried using multiprocessing.Manager and converting the ordered dict to list when passing; multiprocessing.Queue but haven't succeeded so far.

Many thanks for any help!

CompliceU commented 1 year ago

@AngelaZha 不知道你成功了没有呢,我最近也打算尝试一下你的思路,可以交流一下,1215533585,这是我的QQ。我认为可以利用单摄像头进行跟踪,把特征和crop图等信息放到一个具有datbase概念的队列中,在另外的一个进程中进行消费处理,可以是聚类或者其他有效的方法。