🚦 Project Goals

• Apply computer vision to detect and count vehicles per lane.
• Use adaptive scheduling for green/red traffic lights based on vehicle density.
• Export models for edge deployment using ONNX and TensorRT.

🛠 Technologies Used

• YOLOv5s (for fast, lightweight object detection)
• ONNX Runtime

---

🚀 My Contributions

I built upon the original open-source research by:

• 🔧 Improving and optimizing the core algorithm for better vehicle detection accuracy and adaptive timing logic.
• ⚙️ Focusing on CPU and GPU-based inference using ONNX Runtime, enabling deployment on a wider range of hardware without relying on specialized accelerators like TensorRT.
• 📦 Streamlining the model export and deployment workflow, especially for environments where GPU inference is possible but TensorRT is not practical.
• 🧪 Benchmarking performance and ensuring efficient real-time operation on both CPU and GPU setups.

These contributions enhance accessibility, deployment flexibility, and encourage further open-source collaboration.

---

📚 Resources & Tutorials

• YOLOv5: https://github.com/ultralytics/yolov5
• ONNX Export & Optimization: Ultralytics Tutorial Section

---

🤝 Acknowledgements

Originally created by Natnael-k as part of ongoing project. Contributions made to enhance deployment flexibility and performance on standard hardware.

---