Replace Current YOLO Model with Lighter Version for Faster Inference

[SurajSanap]

Files Affected:

• fire_detection.py
• gear_detection.py
• r_zone.py

📌 Current Model Info

• Using ultralytics.YOLO (default checkpoint).
• Confidence threshold: 0.85 (fire & gear), 0.45 (people).
• No optimization applied.

🐞 Problem

The current YOLO model is too heavy for real-time inference on edge devices (e.g., CPU-only machines, Raspberry Pi, Jetson Nano).

• On CPU: FPS drops below 10 FPS, making it unsuitable for live monitoring.
• High memory usage.

✅ Steps to Reproduce

1. Run any detection script (fire_detection.py, gear_detection.py, r_zone.py) on CPU with a live webcam feed.
2. Measure FPS → performance is very low (<10).

💡 Suggested Improvement

• Replace the current YOLO checkpoint with a lighter variant such as:

  • YOLOv8n (nano)
  • YOLOv9-tiny
  • YOLO-Nano / MobileNet-based detectors

• Add a config option so users can choose model size dynamically (n, s, m, l).

📊 Expected Outcome

• Improved FPS (15–30+) on CPU.
• Lower memory usage.
• Contributors can experiment with different models without modifying code.

[dharapandya85]

Please assign me this issue.