Helmet_Detection

This is an Object Detection app built with Streamlit, OpenCV, and the YOLOv10 model. The app allows users to upload an image, choose a pre-trained model, and adjust parameters such as confidence threshold and image size for detection. The processed image with detected objects will be displayed and can be downloaded.

Setup For Using

1. Create new folder

2. Clone yolov10 repo and install requirements

   git clone https://github.com/THU-MIG/yolov10
   cd yolov10
   pip install .

Setup For Training

3. Download dataset and pretrained_model from pretrained_model and dataset

4. Check CUDA version

   nvcc --version

5. Reinstall CUDA 11.7 from CUDA 11.7

6. Modify datasets_dir in settings.yaml at C:\Users\your_user_name\AppData\Roaming\yolov10

Folder Structures

• code: source code
  • predicts: the location to save results during app execution
  • test_results: previously produced results
  • trained_models: 2 models trained by Google Colab and RTX 4050 Ti
  • deploy.py: main app using Streamlit
• safety_helmet_dataset: datasets for training and testing the models
• yolov10: YOLOv10 implementation main directory
• training.py: training sample on Local
• Project.ipynb: training sample on Google Colab

Usage

• Once the environment is configured and dependencies are installed, you can begin using the YOLO model to detect objects in your images.

• To use this app, run this command interminal:

  streamlit run deploy.py

• Sample Image Predictions

  

Recognitions

This project utilizes the YOLOv10 model from THU-MIG