Object Tracking Using DeepSORT and YOLOv5

This project implements object tracking using DeepSORT and YOLOv5. The pipeline combines YOLOv5 for object detection and DeepSORT for object tracking to process video files and generate outputs with bounding boxes and tracking IDs.

Project Structure

├── detect_sort.py # Main script for object detection and tracking

├── YOLOv5_DeepSort_Tracking.ipynb # Jupyter notebook for exploration and visualization

├── yolov5s.pt # Pretrained YOLOv5 weights

├── requirements.txt # Python dependencies

├── graphs.py # Utility functions for bounding box visualization

├── LICENSE # License file (MIT License)

├── pedestrian.mp4 # Input video for demonstration

├── yolov5/ # YOLOv5 model codebase

├── deep_sort_pytorch/ # DeepSORT implementation

├── runs/ # Folder containing output videos and results

├── detect/

-exp/ # Folder where the processed video is saved

├── pycache/ # Python cache files

Features

• Object Detection: Detect objects in videos using YOLOv5.
• Object Tracking: Track detected objects across frames using DeepSORT.
• Output Generation: Save processed videos with bounding boxes and tracking IDs in the runs/detect/exp directory.
• Comparison: View the original video (pedestrian.mp4) and the processed video for a clear understanding of the impact of object tracking.

Installation and Setup

Follow these steps to set up the project and run the object tracking script:

1. Clone the Repository:

   git clone <repository-link>
   cd <repository-folder>

2. Create and Activate Virtual Environment:

   python -m venv .venv
   # Activate the environment:
   # On Windows:
   .venv\Scripts\activate
   # On Linux/Mac:
   source .venv/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Run the Script:

   python detect_sort.py --weights yolov5s.pt --img 640 --source pedestrian.mp4 --view-img

Input and Output

• Input Video:

  • The original video (pedestrian.mp4) demonstrates the scene before applying object detection and tracking.

• Output Video:

  • The processed video, including bounding boxes and tracking IDs, is saved in the runs/detect/exp folder.

Original Video (Before Tracking)	Processed Video (After Tracking)

License

This project is licensed under the MIT License. For more details, see the LICENSE file.

Notes

• Ensure the input video (pedestrian.mp4) is placed in the root directory.
• The detect_sort.py script supports additional configurations like saving detection results as text files, cropping detected objects, and more. Run python detect_sort.py --help for all available options.

Feel free to explore and modify the project for your use cases!