0
Jump to main content

Advances in Deep Learning Towards Fire Emergency Application

Jivitesh Sharma of the Faculty of Engineering and Science at the University of Agder has submitted his thesis entitled “Advances in Deep Learning Towards Fire Emergency Application: Novel Architectures, Techniques and Applications of Neural Networks” and will defend the thesis for the PhD-degree Monday 11 January 2021. (Photo: Private)

The main objective of this thesis is to advance the state-of-the-art methods in Artificial Intelligence and apply them to the important task of Emergency Management. Specifically, we focus on fire emergencies as they are the most common type of hazard.

Jivitesh Sharma

PhD Candidate

The disputation will be held digitally, because of the Corona covid-19-situation. Spectators may follow the disputation digitally – link is available below.

 

Jivitesh Sharma of the Faculty of Engineering and Science at the University of Agder has submitted his thesis entitled “Advances in Deep Learning Towards Fire Emergency Application: Novel Architectures, Techniques and Applications of Neural Networks” and will defend the thesis for the PhD-degree Monday 11 January 2021.

He has followed the PhD-programme at the Faculty of Engineering and Science, with spesialisation in ICT.

Summary of the thesis by Jivitesh Sharma:

Advances in Deep Learning Towards Fire Emergency Application: Novel Architectures, Techniques and Applications of Neural Networks

The main objective of this thesis is to advance the state-of-the-art methods in Artificial Intelligence and apply them to the important task of Emergency Management. Specifically, we focus on fire emergencies as they are the most common type of hazard.

The problem is divided into three stages: Detection, Analysis and Evacuation Planning. We provide Artificial Intelligence based solutions for each stage in a disjoint manner.

Detection:

Current systems lack the high precision of Artificial Intelligence based methods and require more hardware such as smoke detectors, thermal detectors etc.

For detecting an emergency, we use two ways: visual and audio recognition. Novel Artificial Intelligence based computer vision techniques are proposed to detect fire with high accuracy, that might potentially be used with camera systems that we have today.

For audio-based detection, we propose a novel sound recognition model using neural networks that sets the new state-of-the-art in this application. The auditory emergency recognition model is a general-purpose emergency detection system.

Analysis:

We propose the first Artificial Intelligence based emergency analysis tool to thoroughly analyse a disaster-stricken area. A neural network based object detection and segmentation model is proposed. The model distinguishes and segments objects in the emergency environment based on their build material to convey information about the objects’ vulnerability to catch fire. Also, people are also detected and segmented to get a rough head count and location. The model could also provide a rough direction of fire spread.

This can be extremely crucial information for the search and rescue personnel that usually go in without any useful information, to save people. This can potentially reduce the risk of injury or death for the fire fighters and help in rescuing people more efficiently. 

Evacuation:

For the final and probably the most important stage of the emergency management procedure, we propose the first full scale evacuation planning model based on Artificial Intelligence. Specifically, we use Reinforcement Learning to plan optimal evacuation strategies to evacuate all people inside a building. We include various realistic features and scenarios in our simulator including dynamic fire spread, bottleneck (max. number of people inside a room/hallway), uncertainty (to model human behaviour), timeliness constraint (to evacuate everyone in the least amount of time), multiple fires etc.

We train a Reinforcement Learning agent on this simulator with the objective of evacuating everyone in the least amount of time while avoiding any hazardous areas in the building. In order to scale our model to work on large buildings, we employ attention based Reinforcement Learning. We show the effectiveness of our method by running a simulation for fire evacuation on our own UiA building in Grimstad. Our method is able to evacuate nearly 1000 people from the UiA building in nearly optimal time, without putting anyone in harm’s way. We also provide mathematical guarantees for our model.

 

Disputation facts:

The trial lecture and the public defence will take place online, via the Zoom conferencing app (link below)

Associate Professor Morgan Konnestad of the Department of Information and Communication Technology, Faculty of Engineering and Science, UiA, will chair the disputation.

The trial lecture at 14:15 hours
Public defence at 16:00 hours

 

Given topic for trial lecture«Ensembles and the bias / variance dilemma»

Thesis Title: “Advances in Deep Learning Towards Fire Emergency Application: Novel Architectures, Techniques and Applications of Neural Networks

Search for the thesis in AURA - Agder University Research Archive, a digital archive of scientific papers, theses and dissertations from the academic staff and students at the University of Agder.

The thesis is available here:

https://uia.brage.unit.no/uia-xmlui/handle/11250/2721827

or

 

The CandidateJivitesh Sharma (1991, New Delhi, India) Bachelor of Technology from Guru Gobind Singh Indraprastha University, New Delhi, India (2013), Masters degree in Technology from Maulana Azad National Institute of Technology, Bhopal, India (2016).

Opponents:

First opponent: Associate Professor Paul W. Munro, School of Computing and Information, University of Pittsburgh, USA

Second opponent: Professor Heri Ramampiaro, Department of Computer Science, NTNU

Professor Frank Reichert, Department of Information and Communication Technology, University of Agder, is appointed as the administrator for the assessment commitee.

Supervisors were Professor Ole-Christoffer Granmo, UiA (main supervisor) and Professor Morten Goodwin, UiA (co-supervisor)

 

What to do as an audience member:

The disputation is open to the public, but to follow the trial lecture and the public defence, which is transmitted via the Zoom conferencing app, you have to register as an audience member:

https://uiano.zoom.us/meeting/register/u5UvcOmqrzIrGdL04nVc2EKJ_1KiDFhn_SCO

A Zoom-link will be returned to you.

(Here are introductions for how to use Zoom: support.zoom.us if you cannot join by clicking on the link.)

We ask audience members to join the virtual trial lecture at 14:05 at the earliest and the public defense at 15:50 at the earliest. After these times, you can leave and rejoin the meeting at any time. Further, we ask audience members to turn off their microphone and camera and keep them turned off throughout the event. You do this at the bottom left of the image when in Zoom. We recommend you use ‘Speaker view’. You select that at the top right corner of the video window when in Zoom.

Opponent ex auditorio:

The chair invites members of the public to pose questions ex auditorio in the introduction to the public defense, with deadlines. It is a prerequisite that the opponent has read the thesis. Questions can be submitted to the chair Morgan Konnestad at e-mail morgan.konnestad@uia.no