Fire Detection and alarming using Deep Learning and Edge computing

Aim of this project is to develop a deep learning model to detect early fire on a edge device for laser machine applications.

It has been achieved through below steps.

1. Kaggle online dataset is used for training.
2. Engineered fire features using OpenCV APIs
3. Extended VGG16 model and retrained with fire dataset
4. Evaluated and tested the model performance on PC
5. Compressed model size by quantization technique
6. Converted model to TFLite model and deployed it on Raaspberry PI3 board.
7. Achieved inference performance of 20f/minute on RPi3 board.

Results:

Fire Detection on PC

https://drive.google.com/file/d/1nwjUtzd-wKx38WXLqq1rBISrr6xdbe-G/view?usp=sharing

Fire Classification on RPI3 board

https://drive.google.com/file/d/17ET_BHGeVueDgxhbbx-6rUoU-Xr2OSCb/view?usp=sharing

Please check report for more information.