InfernoGuard is an AI-driven fire detection system built using YOLOv5, optimized for real-time thermal camera-based fire detection. It utilizes ONNX, TensorFlow Lite (TFLite), and TensorFlow.js (TF.js) for seamless deployment across edge devices, cloud servers, and web applications.

By leveraging OpenCV’s Deep Neural Network (DNN) module, InfernoGuard can detect fire outbreaks in images, videos, and live streams, making it an ideal solution for early fire detection in industrial plants, forests, warehouses, and residential buildings.

✅ Thermal Camera-Based Fire Detection – Works in low visibility conditions.
✅ Optimized for Multiple Platforms:

• ONNX (High-performance inference on CPU/GPU)
• TFLite (Optimized for mobile, edge, and embedded systems)
• TF.js (Runs in a browser for web-based monitoring)
  ✅ Real-Time Performance – Works with CCTV, drones, and IoT-enabled cameras.
  ✅ Edge AI Deployment – Compatible with Raspberry Pi, NVIDIA Jetson, and cloud-based solutions.

🔹 Early fire detection in industrial zones, warehouses, and factories.
🔹 Forest fire monitoring using drones.
🔹 Smart surveillance for residential and commercial buildings.
🔹 IoT integration for automatic fire alarm activation.

import cv2
from yolo_predictions import YOLO_Pred

# Load Model
yolo = YOLO_Pred('./Model/weights/best.onnx', 'data.yaml')

# Load Image
img = cv2.imread('test_image.jpg')

# Perform Detection
img_pred = yolo.predictions(img)

# Display Results
cv2.imshow('Fire Detection', img_pred)
cv2.waitKey(0)
cv2.destroyAllWindows()

cap = cv2.VideoCapture(0)  # Use webcam

while True:
    ret, frame = cap.read()
    if not ret:
        break

    pred_image = yolo.predictions(frame)
    cv2.imshow('Fire Detection', pred_image)

    if cv2.waitKey(1) == 27:  # Press 'ESC' to exit
        break

cap.release()
cv2.destroyAllWindows()

📈 Training Metrics (Loss, Accuracy, Confusion Matrix) are available in the results/ folder.

├── Model/
│   ├── weights/
│   │   ├── best.onnx  # ONNX Model
│   │   ├── best.tflite  # TFLite Model
│   │   ├── best_web_model/  # TF.js Model
│   ├── results/  # Training graphs and evaluation results
│
├── predictions/
│   ├── detect.py  # Image/Video/Webcam detection script
│   ├── yolo_predictions.py  # YOLO inference class
│   ├── utils.py  # Helper functions
│
├── dataset/
│   ├── images/
│   ├── labels/
│── data.yaml  # Dataset configuration
│
├── README.md  # Project Documentation

git clone https://github.com/83Gh0st/InfernoGuard.git
cd InfernoGuard

🔹 On Images

python3 detect.py --source test_image.jpg

🔹 On Webcam

python3 detect.py --source 0

import torch

model = torch.load('best.pt', map_location='cpu')
dummy_input = torch.randn(1, 3, 640, 640)
torch.onnx.export(model, dummy_input, "best.onnx", opset_version=11)

onnx-tf convert -i best.onnx -o best.pb
tflite_convert --saved_model_dir=best.pb --output_file=best.tflite

tensorflowjs_converter --input_format=tf_saved_model best.pb best_web_model/

1️⃣ Copy the best_web_model/ to your web server.
2️⃣ Load the model in JavaScript:

const model = await tf.loadGraphModel('best_web_model/model.json');
const img = tf.browser.fromPixels(document.getElementById('input_image'));
const predictions = model.predict(img);

🔹 IoT Integration for automated fire alarm triggers.
🔹 Deploy as a cloud-based AI API for smart surveillance.
🔹 Expand dataset to include more fire intensity variations.
🔹 Optimize for real-time edge AI applications (e.g., NVIDIA Jetson Nano).

This project is licensed under the MIT License – Free to use, modify, and distribute.

Developed by @83Gh0st 🔥
💬 Contact: GitHub

🔥 Star this repo if you found it useful! ⭐