Open changtimwu opened 2 years ago
more paper survey https://zhuanlan.zhihu.com/p/393663819
some discussion https://github.com/openvinotoolkit/open_model_zoo/issues/1497
_The main algorithm of the multi_camera_multi_targettracking 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.
https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/multi_camera_multi_target_tracking_demo/python/mc_tracker
https://github.com/openvinotoolkit/open_model_zoo/issues/2240