FireProof

An innovative way to detect fire and thus execute safety protocol for residents in a buliding.

Table of Contents

• Problem Statement
• Solution
• Machine Learning Model
• How to run
• Demo
• Project Links
• Team Members

Problem Statement

Every day, about 60 people die due to a fire hazard in India. It causes countless lives, mostly due to late detection of fire and people not being able to find a safe route to escape the burning buildings. The excess amount of smoke and gas in case of fire cause difficult visibility, rendering all escape plans useless.

Our solution

A web app which takes images from various CCTV in building cameras runs a Neural Net and incase of a fire sends an alert and suggests best possible escape route to users. Right now we have demonstrated on images but since the model is small enough and the latency is low, it can be used with live video stream.

The path in general

ML model

1. We used data which consisted of images from CCTV to make it as real as possible.
2. We have used MobileNetV2 as the base model and achieved 96% accuracy on train and 95% on test with 4 fold train-test data split.
3. With the help OpenCV tools we measure the percent area recorded by camera that is affected by fire and assign a weight to that node.
4. Training and validation Accuracy and Loss graph

Instruction

1. Clone the project

cd /path/to/directory

2. Install all the dependencies

pip install -r requirements.txt

3. Save photos in images folder
4. execute while in the directory

python app.py

Demo

The landing page
Path if there is a fire
If no fire is detected

Important Links

Presentation

Team

Harsh Kumar Singh
Sonali Verma
Somya Jain
Surya Prakash Mishra