README.md

Yolov4-deepsort-helmet-detection

helmet(hard hat) detector trained with darknet framework YOLOv4 model.

tested environment

• Windows 10 x64 2020 (build 19041.388)
• NVIDIA RTX 2070 Super
• CUDA 10.1 / CuDNN 7.6.5
• Python 3.7.7 x64
• tensorflow 2.2.0 GPU

---

trained weight

• Google Drive
  • put weight file in ./configs
• To train with your own datasets, You should do yourself with darknet. And need to change some parameters

used dataset

• roboflow.ai public dataset + about 100 images

---

dependencies

• python
  • opencv-python, numpy, scikit-image
    • to draw boxes and texts on image
  • tensorflow 2.2.0
    • to track object with DeepSORT model
  • matplotlib
    • to create color map
• CUDA 10.1 / CuDNN 7.6.5
• darknet
  • for yolov4 object detection
  • dark.dll, pthreadVC2.dll (windows)
    • Pre-Compiled (I'm not sure if it works on different machine)- Google Drive
  • libdarknet.so (linux)

prepare environment and run

pip install -r requirements.txt

python detectandtrack.py

you should modify code to make change

example

citation

https://github.com/AlexeyAB/darknet

https://github.com/theAIGuysCode/yolov3_deepsort

https://public.roboflow.ai/object-detection/hard-hat-workers