people tracking methodology

[changtimwu]

https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/multi_camera_multi_target_tracking_demo/python/mc_tracker
openvinotoolkit/open_model_zoo#2240

[changtimwu]

https://arxiv.org/abs/1906.01357
https://blog.csdn.net/weixin_40194996/article/details/105589193

[changtimwu]

more paper survey
https://zhuanlan.zhihu.com/p/393663819

https://zhuanlan.zhihu.com/p/394007579

[changtimwu]

some discussion
openvinotoolkit/open_model_zoo#1497

The main algorithm of the multi_camera_multi_target_tracking is based on a re-identification model which extracts a re-id feature from every bounding box. Also there is used additional heuristics such as IoU between bounding boxes, velocity constraint e.t.c. Finally we did not write an article, we used already known methods. We were focused on re-identification models in this algorithm.

No, it is not a version of deep sort tracker but some ideas probably were borrowed from this method.