Target Tracker

Real-time object detection using YOLO11/YOLOv8 for webcams and video files. Optimized for NVIDIA GPUs with CUDA support and automatic CPU fallback.

Features

• YOLO11 Support: Latest YOLO11 models with 10-15% better performance than YOLOv8
• GPU Accelerated: Optimized for NVIDIA GPUs with CUDA 11.8
• Real-time Detection: Webcam and video file support
• Interactive Controls: Switch models and resolutions during runtime
• Threaded Processing: 30-40% performance boost with parallel frame reading
• 10 Models Available: 5 YOLO11 models + 5 YOLOv8 models (backward compatible)

Quick Start

Prerequisites

• Python 3.11+
• NVIDIA GPU with CUDA 11.8 (optional, auto-falls back to CPU)
• Webcam (for live detection)

Installation

1. Clone the repository

git clone https://github.com/suteny0r/Target-Tracker.git
cd Target-Tracker

2. Create and activate virtual environment

# Windows
python -m venv venv
venv\Scripts\activate

# Linux/Mac
python -m venv venv
source venv/bin/activate

3. Install PyTorch with CUDA (for GPU acceleration)

pip install torch==2.5.1+cu118 torchaudio==2.5.1+cu118 torchvision==0.20.1+cu118 --index-url https://download.pytorch.org/whl/cu118

For CPU-only:

pip install torch torchvision torchaudio

4. Install dependencies

pip install -r requirements.txt

Usage

Basic webcam detection:

python object_detection.py

Process video file:

python object_detection.py --video path/to/video.mp4

Threaded mode (30-40% faster):

python object_detection.py --threaded

GPU Benchmark mode:

python object_detection.py --no-display --threaded

Interactive Controls

While running, use these keys:

• W / S - Switch to larger/smaller model (10 models available)
• A / D - Decrease/increase display resolution (480p/720p/1080p)
• Q - Quit

Available Models

YOLO11 (Current, Recommended)

• yolo11n - Nano (fastest, ~100+ FPS on RTX 3090)
• yolo11s - Small (default on CPU)
• yolo11m - Medium
• yolo11l - Large
• yolo11x - Extra Large (default on GPU, most accurate)

YOLOv8 (Legacy, Still Supported)

• yolov8n through yolov8x (same size variants)

Performance: YOLO11 models are ~10-15% faster and ~5-10% more accurate than YOLOv8.

Note: Models auto-download from Ultralytics on first use (~6MB to 131MB each).

Command-Line Options

--camera-index 0        Select webcam (default: 0)
--video PATH           Path to video file
--imgsz 640            Model input resolution (default: 640)
--conf 0.5             Confidence threshold (default: 0.5)
--threaded             Enable threaded frame reading (30-40% faster)
--no-display           Benchmark mode without display
--force-cpu            Force CPU mode (disable GPU)

Performance

On NVIDIA RTX 3090 with YOLO11x:

• Threaded mode: ~60-80 FPS (1080p display)
• Benchmark mode: ~100+ FPS (no display overhead)
• CPU mode: ~10-15 FPS

Building Standalone Executable

pyinstaller object_detection.spec

Output: dist/object_detection.exe

Project Structure

Target-Tracker/
├── object_detection.py      # Main application
├── benchmark_pure_gpu.py   # GPU benchmark script
├── requirements.txt        # Python dependencies
├── CLAUDE.md              # Developer documentation
├── README.md              # This file
└── venv/                  # Virtual environment (not in git)

Troubleshooting

CUDA not detected:

• Verify NVIDIA drivers are installed: nvidia-smi
• Check CUDA 11.8 compatibility with your GPU
• Fallback to CPU mode with --force-cpu

Low FPS:

• Use --threaded flag
• Switch to smaller model (press S)
• Reduce display resolution (press A)
• Lower --imgsz value

Models not downloading:

• Check internet connection
• Ultralytics will auto-download on first use
• Manual download: https://docs.ultralytics.com/

Version History

• v1.1.0 (2025-10-20): YOLO11 upgrade, 10 models available, performance improvements
• v1.0.0 (2025-10-20): Initial release with YOLOv8

License

See repository for license details.

Acknowledgments

• Ultralytics YOLO - YOLO11 and YOLOv8 implementation
• PyTorch - Deep learning framework
• OpenCV - Computer vision library