Deep sort

yolo5 pytorch
https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch (uses https://github.com/NirAharon/BOT-SORT) (conda install instructions(oct 2021,outdated a bit) at https://pastebin.com/tQKGWbym and https://github.com/KaiyangZhou/deep-person-reid) This repository contains a moded version of PyTorch YOLOv5 (https://github.com/ultralytics/yolov5). It and tracks up to 79 different objects. See https://github.com/PaddlePaddle/PaddleDetection, https://github.com/masterbin-iiau/unicorn as listed by https://paperswithcode.com/task/multiple-object-tracking for alternative to brostrom. https://github.com/mikel-brostrom/Yolov5_StrongSORT_OSNet/issues/594 Multi camera tracking to work with a unified understanding among all the cameras https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/368 (solved oct.2022). Kalman prediction of next position of the object, use with sentry turrent and net gun. A net gun means you can have swarming bots in between people and on accidental firing(which will happen) they have a harmless net over them, obviously not possible with pepperball mounted turret. If though you do insist, then at least have the bots locked in behind steel cages with a siren and flash orange light on each, with the door opened by a Linear actuators, giving you some escape time. https://github.com/mikel-brostrom/Yolov5_DeepSort_OSNet/issues/365 Yeah, divide the vx and vy by (t5-t1) to get the velocities in pixel per second, but only for the real time video streaming. offline videos will be processed every frame as fast as possible so it won't make sense to use the timings here to get pixel per second, it will still be pixel per frame. There is no way to get meter per second, since meters is dependent on scene and lens and objects distance from camera... Use lidar, depthperception and try a jabelone type neural training for a fixed lens and area to inver distance in meters.

Note that this is sort of a hackish way to get the units you want, ie pixel per second. Probably need to revisit the way kalmanfilter is used in the code to really get the entire tracking portion of the code to perform the way it is meant to. i.e. estimate both the state and covariances of it's position and velocity in pixel and pixel per seconds.

zqpei
https://github.com/ZQPei/deep_sort_pytorch

yolo4
https://github.com/LeonLok/Deep-SORT-YOLOv4 I swapped out YOLO v3 for YOLO v4 and added the option for asynchronous processing, which significantly improves the FPS. However, FPS monitoring is disabled when asynchronous processing is used since it isn't accurate. https://github.com/mkocabas/multi-person-tracker uses SORT for multiperson tracking. https://github.com/JunweiLiang/Object_Detection_Tracking We utilize state-of-the-art object detection and tracking algorithm in surveillance videos. Our best object detection model basically uses Faster RCNN with a backbone of Resnet-101 with dilated CNN and FPN. The tracking algo (Deep SORT) uses ROI features from the object detection model. The ActEV trained models are good for small object detection in outdoor scenes. For indoor cameras, COCO trained models are better https://github.com/ZQPei/deep_sort_pytorch This is an implement of MOT(Multiple Object Tracking) tracking algorithm deep sort. Deep sort is basicly the same with sort but added a CNN model to extract features in image of human part bounded by a detector. This CNN model is indeed a RE-ID model and the detector used in PAPER is FasterRCNN, and the original source code is HERE. However in original code, the CNN model is implemented with tensorflow, which I'm not familier with. SO I re-implemented the CNN feature extraction model with PyTorch, and changed the CNN model a little bit. Also, I use YOLOv3 to generate bboxes instead of FasterRCNN.

yolo deepsort
https://github.com/Qidian213/deep_sort_yolov3 combine with Pose estimation

yolo4 version
https://github.com/LeonLok/Deep-SORT-YOLOv4

nwojke
https://github.com/nwojke/deep_sort https://github.com/louxy126/deep_sort_yolov3 is based on nwojke and a download link to the original weights from the nwojke repo for tracking people was removed by nwojke. The weights allows for training to track any object. https://github.com/bendidi/Tracking-with-darkflow direct download link of orignal nwojke weights as provided by louxy126. file is resources.zip from https://towardsdatascience.com/people-tracking-using-deep-learning-5c90d43774be

links
Object tracking yolo SORT tracking