A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Smart Fire Safety: Real-Time Segmentation and Alerts Using Deep Learning
Fires are the major causes of property damage, injuries, and death worldwide. The ability to avoid or reduce the effects of fires depends on their early identification. The accuracy and responsiveness of conventional fire detection systems, such as smoke detectors and heat sensors, are constrained. Computer vision-based fire and smoke detection systems have been suggested as a replacement for conventional systems in recent years. To tackle the challenges a robust real-time framework has been proposed, whereby, images are taken from cameras and using a custom train YOLOv8 object segmentation model smoke and fires are localized in the image which are then fed to an expert system for alert generation. The expert system makes decisions on the fire status based on its size and growth across multiple frames. Furthermore, A new dataset was meticulously curated and annotated for the segmentation task, to assess the efficacy of the proposed system, comprehensive benchmarking was conducted on the proposed dataset using a suite of benchmarks. The proposed system achieved an mAP score of 74.9% on the benchmark dataset. Furthermore, it was observed that employing segmentation for localization as opposed to detection, resulted in system accuracy improvement. The system can immediately identify fires and smoke and send accurate alerts to emergency services.
Smart Fire Safety: Real-Time Segmentation and Alerts Using Deep Learning
Fires are the major causes of property damage, injuries, and death worldwide. The ability to avoid or reduce the effects of fires depends on their early identification. The accuracy and responsiveness of conventional fire detection systems, such as smoke detectors and heat sensors, are constrained. Computer vision-based fire and smoke detection systems have been suggested as a replacement for conventional systems in recent years. To tackle the challenges a robust real-time framework has been proposed, whereby, images are taken from cameras and using a custom train YOLOv8 object segmentation model smoke and fires are localized in the image which are then fed to an expert system for alert generation. The expert system makes decisions on the fire status based on its size and growth across multiple frames. Furthermore, A new dataset was meticulously curated and annotated for the segmentation task, to assess the efficacy of the proposed system, comprehensive benchmarking was conducted on the proposed dataset using a suite of benchmarks. The proposed system achieved an mAP score of 74.9% on the benchmark dataset. Furthermore, it was observed that employing segmentation for localization as opposed to detection, resulted in system accuracy improvement. The system can immediately identify fires and smoke and send accurate alerts to emergency services.
Smart Fire Safety: Real-Time Segmentation and Alerts Using Deep Learning
Farhan Khan (author) / Sarmad Rafique (author) / Salman Khan (author) / Laiq Hasan (author)
2024-05-22
International Journal of Innovations in Science & Technology; Vol. 6 No. 5 (2024): ICTIS Spl Issue; 105-115 ; 2709-6130 ; 2618-1630
Article (Journal)
Electronic Resource
English
DDC:
624
SafeSense: Real-time Safety Alerts for Construction workers
| TIBKAT | 2024
Deep Learning-Based Real-Time Crack Segmentation for Pavement Images
| Springer Verlag | 2021
Testing Real-Time In-Home Fall Alerts with Embedded Depth Video Hyperlink
| British Library Conference Proceedings | 2015
Testing Real-Time In-Home Fall Alerts with Embedded Depth Video Hyperlink
| Springer Verlag | 2014