search

John1liu / YOLOV5-DeepSORT-Vehicle-Tracking-Master

In this project, urban traffic videos are collected from the middle section of Xi 'an South Second Ring Road with a large traffic flow, and interval frames are extracted from the videos to produce data sets for training and verification of YOLO V5 neural network. Combined with the detection results, the open-source vehicle depth model data set is used to train the vehicle depth feature weight file, and the deep-sort algorithm is used to complete the target tracking, which can realize real-time and relatively accurate multi-target recognition and tracking of moving vehicles.
77 stars 10 forks source link

readme

YOLOV5-DeepSORT-Vehicle-Tracking-Master

In this project, urban traffic videos are collected from the middle section of Xi 'an South Second Ring Road with a large traffic flow, and interval frames are extracted from the videos to produce data sets for training and verification of YOLO V5 neural network. Combined with the detection results, the open-source vehicle depth model data set is used to train the vehicle depth feature weight file, and the deep-sort algorithm is used to complete the target tracking, which can realize real-time and relatively accurate multi-target recognition and tracking of moving vehicles. And I also upload my dissertation of bachelor degree, I think you can learn about how to set up this project via this paper (But it is in simplified Chinese.).

Installation

Python 3.6 or later with all requirements.txt dependencies installed, including torch>=1.7. To install run:

pip install -r requirements.txt

Run Tracker

After you download this project, please download the weight of YOLO V5 model and Deep-SORT model respectively. You can download the YOLO weight trained by me from https://drive.google.com/file/d/1-8Xm3eUMMJF5XNiF649kqnqoYeWhv3kT/view?usp=sharing or choose to download the pretrained weight of the YOLO V5 model with using the ./yolov5/weights/downloadweight.sh.

You can download the Deep-SORT weight trained by me from https://drive.google.com/file/d/1-GjB1pGudjM70C1Jk7WhWUWShzCvvvra/view?usp=sharing or choose to download the pretrained weight of the Deep-SORT model from https://drive.google.com/file/d/1-EsIHxEqr9elRryUqPBcBBAY3qoIEbZS/view?usp=sharing . And then put the weight into the folder namely /deep_sort_pytorch/deep_sort/deep/checkpoint.

Finanlly, you can choose to download the test video producted by me from https://drive.google.com/file/d/1cMch88P8xZ95SoHN88aC53oiOLlD3Egb/view?usp=sharing.

Tracking can be run on most video formats

python3 track.py --source ...

MOT compliant results can be saved to inference/output by 

python3 track.py --source ... --save-txt

Result

https://drive.google.com/file/d/1-2hL9UE3i24bRE0an9HTDbN5bJBEeXfg/view?usp=sharing 1

Train YOLO weights with your datasets

Please have a reference on https://github.com/ultralytics/yolov5.

Train Deep-SORT weights with your datasets

I have update my datasets about the appearance of cars from https://drive.google.com/file/d/1lushuv4QMTmfFwURU1Ug6mezkzQAMi0S/view?usp=sharing. You can choose to read the original paper of Deep-SORT to learn how to train it. I will also update quickly. Updating now...... I consider to create a new repository about it now......

References

https://github.com/ultralytics/yolov5 Simple Online and Realtime Tracking with a Deep Association Metric https://arxiv.org/abs/1703.07402