multi-tracking-YOLOv8-ByteTrack

Cows/Animal Multi-Tracking with YOLOv8 and ByteTrack

Demo

This demo is a simple software that I created using my studies and courses. Quem eu sou: https://luiz-amaral-portfolio.vercel.app/

Overview

This software is designed for real-time object detection and tracking in video streams, using YOLO (You Only Look Once) for detection and ByteTrack for multi-object tracking. It is suitable for tracking objects in video footage, and can run on both GPU and CPU.

Key Features:

• Real-time video reading and frame buffering in a dedicated thread.
• Asynchronous frame processing (detection + tracking) in a separate thread.
• Uses YOLOv8 for object detection and ByteTrack for robust multi-object tracking.
• Visualizes both detection and tracking results side-by-side.
• Supports both GPU and CPU execution.

---

System Architecture

The system is built around two main threads:

1. Video Reading Thread

   • Reads frames from a video file.
   • Stores the latest frame in a thread-safe singleton buffer.

2. Processing Thread

   • Waits for new frames.
   • Runs YOLO object detection.
   • Passes detections to ByteTrack for tracking.
   • Draws detection and tracking results.
   • Combines and displays the results.

Diagram

+-------------------+         +---------------------+         +---------------------+
|                   |         |                     |         |                     |
|  Video File/Camera| ---->   | Video Reading Thread| ---->   |  Frame Buffer       |
|                   |         | (camera_video.py)   |         | (frames.py)         |
+-------------------+         +---------------------+         +---------------------+
                                                                      |
                                                                      v
                                                        +--------------------------+
                                                        | Processing Thread        |
                                                        | (processador.py)        |
                                                        | - YOLO Detection        |
                                                        | - ByteTrack Tracking    |
                                                        | - Visualization         |
                                                        +--------------------------+
                                                                      |
                                                                      v
                                                        +--------------------------+
                                                        |   Display/Output         |
                                                        +--------------------------+

---

Example Output

Below is an example of the software in action. It shows ByteTrack tracking results (green boxes with IDs):

How It Works

1. Video Reading (camera/camera_video.py)

• A thread reads frames from a video file.
• Each frame is stored in a singleton buffer (Frames class).
• The reading thread can be stopped when the video ends or by user command.

2. Frame Buffer (camera/frames.py)

• Implements a thread-safe singleton to store and retrieve the latest frame.
• Notifies the processing thread when a new frame is available.

3. Processing (core/processador.py)

• A separate thread waits for new frames.
• Runs YOLOv8 for object detection (configurable for GPU/CPU).
• Passes detections to ByteTrack for multi-object tracking.
• Draws bounding boxes and tracking IDs.
• Combines detection and tracking results for display.

4. Visualization (core/funcoes.py)

• Draws detection boxes (blue) and tracking boxes/IDs (green).
• Shows object trajectories with colored lines.
• Combines detection and tracking images side-by-side.

---

Running the Software

1. Install dependencies (Python, OpenCV, Ultralytics YOLO, ByteTrack, etc.)
2. Place your video file in the appropriate directory (default: ./apoio/cows3.mp4).
3. Run main.py:

   python main.py
   

4. Press q to quit the visualization window.

---

GPU/CPU Support

• The software automatically detects and uses GPU if available (see utils.obter_cpu_gpu()).
• Falls back to CPU if no GPU is detected.

---

File Structure

camera/
  camera_video.py   # Video reading thread
  frames.py         # Frame buffer (singleton)
core/
  processador.py    # Processing thread (YOLO + ByteTrack)
  funcoes.py        # Helper functions for drawing, etc.
main.py             # Entry point

---

---

License

MIT License