Efficient and Compact Convolutional Neural Network Architectures for Non-temporal Real-time Fire Detection

TL;DR

This paper introduces two efficient CNN architectures, NasNet-A-OnFire and ShuffleNetV2-OnFire, optimized for real-time fire detection in videos, achieving high accuracy and speed on low-powered devices.

Contribution

The paper proposes two novel compact CNN architectures specifically designed for non-temporal real-time fire detection, improving accuracy and computational efficiency over existing methods.

Findings

Achieved 95% accuracy for full-frame binary fire classification.

Real-time processing at 40 fps and 18 fps for different tasks.

Successfully deployed on low-powered Nvidia Xavier-NX device at 49 fps.

Abstract

Automatic visual fire detection is used to complement traditional fire detection sensor systems (smoke/heat). In this work, we investigate different Convolutional Neural Network (CNN) architectures and their variants for the non-temporal real-time bounds detection of fire pixel regions in video (or still) imagery. Two reduced complexity compact CNN architectures (NasNet-A-OnFire and ShuffleNetV2-OnFire) are proposed through experimental analysis to optimise the computational efficiency for this task. The results improve upon the current state-of-the-art solution for fire detection, achieving an accuracy of 95% for full-frame binary classification and 97% for superpixel localisation. We notably achieve a classification speed up by a factor of 2.3x for binary classification and 1.3x for superpixel localisation, with runtime of 40 fps and 18 fps respectively, outperforming prior work in the field presenting an efficient, robust and real-time solution for fire region detection. Subsequent implementation on low-powered devices (Nvidia Xavier-NX, achieving 49 fps for full-frame classification via ShuffleNetV2-OnFire) demonstrates our architectures are suitable for various real-world deployment applications.