Car Tracking Algorithms

Implementation and comparison of different car tracking algorithms: SORT, DeepSORT, ByteTrack, and HybridTrack.

Overview

This project provides a comprehensive framework for tracking vehicles in video streams using state-of-the-art tracking algorithms. It includes implementations of SORT, DeepSORT, ByteTrack, and a custom HybridTrack algorithm.

Latest Performance Results

Tracker Comparison (1500 frames)

Tracker	FPS	Tracks/Frame	Detection (ms)	Tracking (ms)	MOTA (%)	ID Switches
ByteTrack	6.45	7.36	151.45	3.54	83	46
DeepSORT	3.59	5.08	141.25	137.00	79	75
SORT	34.87	4.62	17.49	11.19	65	137
HybridTrack	19.53	4.78	49.84	1.38	94	15

Key Findings

• HybridTrack achieves balanced performance between speed and accuracy
• Improved track consistency with enhanced filtering
• Better handling of occlusions and ID switches

Recent Updates

• ✅ Added multi-tracker comparison visualization
• ✅ Implemented real-time performance metrics
• ✅ Enhanced tracker initialization
• ✅ Improved visualization system
• ✅ Added comprehensive results analysis

Features

• ✅ Three tracking algorithms implemented:
  • SORT (Simple Online and Realtime Tracking)
  • DeepSORT (Deep Learning enhanced SORT)
  • ByteTrack (State-of-the-art tracking)
• ✅ Multiple detector options:
  • Simple detector (background subtraction)
  • YOLOv8 detector
  • ByteYOLO detector (optimized for ByteTrack)
• ✅ Real-time visualization
• ✅ Configurable parameters via YAML
• ✅ Interactive demo application

Installation

1. Clone the repository:

git clone https://github.com/yourusername/Car_Tracking_Algorithms.git
cd Car_Tracking_Algorithms

2. Install dependencies:

pip install -r requirements.txt
pip install ultralytics  # For YOLOv8 detector

Project Setup

1. Create required directories:

mkdir -p data/datasets
mkdir -p data/models
mkdir -p data/outputs

2. Add video files:

• Place your video files in data/datasets/
• Supported formats: .avi, .mp4

3. Verify configurations:

• Check that config files exist in configs/
• Verify parameters in config files

Running the Demo

1. From project root directory:

python demo/demo.py

2. Follow the interactive prompts to:

• Select tracking configuration (SORT, DeepSORT, or ByteTrack)
• Select input video

3. View results:

• Real-time tracking visualization will be shown
• Processed video will be saved to data/outputs/
• Press 'q' to quit the demo

Tracking Methods

SORT

• Simple and fast tracking
• Uses Kalman filter for motion prediction
• IoU-based matching
• Best for simple scenarios with clear visibility

DeepSORT

• Enhanced tracking with deep learning features
• Better identity preservation
• Uses YOLOv8 for detection
• Good balance of speed and accuracy

ByteTrack

• State-of-the-art performance
• Handles low-confidence detections
• Advanced state estimation
• Best for complex scenarios

HybridTrack

• Based on ByteTrack's proven framework
• Enhanced track management system
• Efficient Kalman filtering
• IoU-based matching with Hungarian algorithm
• Good balance between speed and accuracy
• Suitable for real-time applications with moderate complexity

Configuration

Each tracker has its own configuration file in configs/:

• sort_config.yaml
• deep_sort_config.yaml
• byte_track_config.yaml

Key parameters can be adjusted in these files.

Project Structure

Car_Tracking_Algorithms/
├── configs/           # Configuration files
├── data/             # Data storage
│   ├── datasets/     # Input videos
│   ├── models/       # Model weights
│   └── outputs/      # Processed videos
├── demo/             # Demo application
├── src/              # Source code
│   ├── trackers/     # Tracker implementations
│   └── utils/        # Shared utilities
└── tests/            # Test suite

Troubleshooting

If you encounter issues:

1. Verify all dependencies are installed
2. Check video file exists in data/datasets
3. Ensure you're running from project root
4. Verify config files exist in configs directory
5. Check that data/outputs directory exists for saving results

Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Create a new Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

HybridTrack: A Unified Approach

Our HybridTrack algorithm successfully combines the strengths of three leading tracking methods:

Integration of Methods

1. From SORT:

   • Efficient Kalman filter for motion prediction
   • Hungarian algorithm for optimal assignment
   • Fast IoU-based matching

2. From DeepSORT:

   • Enhanced track management system
   • Robust track validation criteria
   • Hit streak based confirmation

3. From ByteTrack:

   • Dual-threshold detection handling
   • Advanced state estimation
   • Track recovery mechanism

Key Innovations

• Sequential ID Management:

  • Strictly increasing ID assignment
  • No ID recycling to prevent confusion
  • Continuous tracking history preservation

• Stable Track Maintenance:

  • Combined motion and IoU matching
  • Adaptive track validation criteria
  • Enhanced state prediction

• Robust Detection Association:

  • Two-stage matching strategy
  • High/low confidence detection handling
  • Improved occlusion handling

Performance Highlights

• Maintains consistent tracking through occlusions
• Sequential ID assignment (1,2,3...)
• Stable bounding boxes
• Efficient computation (~20 FPS)

Contributing

1. Fork the repository.
2. Create your feature branch.
3. Commit your changes.
4. Push to the branch.
5. Create a new Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.