Smart Fire Detection using Pi and AWS

Smart Fire Detection using Pi and AWS

• 1.
  Smart Fire Alarm UsingRaspberry Pi and AWS Advanced Distributed Systems Project
• 2.
  Outline Introduction Background and RelatedWork Problem Statement Proposed Solution Architecture Raspberry Pi Amazon Web Services(AWS) MQTT Protocol Demo Result, Conclusion and Future work
• 3.
  Introduction • Safety andsecurity have always been the prime concern of mankind. • With so many advancements in the fields of machine learning, computer vision, cloud infrastructure, it provides us with an opportunity to leverage these techniques to address the concerns. • We want to tap the potentials of cloud infrastructures and machine learning to detect the fire and implement the smart fire detection system so that we could alert the owner/residents as early as possible to minimize the casualties.
• 4.
  Background and RelatedWork • Fires are very common things happening in the world. Every 24 seconds, a fire department in the United States responds to a fire somewhere in the nation. (Ref : Insurance Information Institute)[5] • The thermal imaging approach, has lots of drawbacks like high initial cost, images are difficult to interpret in specific objects having erratic temperatures, etc. [6] • In the earlier research, researcher has used raspberry pi with Arduino board for smart fire alarms[1]
• 5.
  Problem Statement To build firealert internet of things (IoT) system using Raspberry pi and cloud (AWS) To reduce false positive alarms and provide adequate alerts to the users
• 6.
  Proposed Solution • Weproposed the architecture in alignment with our aim, by using raspberry pi and Amazon Web Services (Cloud) and further embedding the alert system in the smart home hub like Amazon Alexa or Google Home
• 7.
  Architecture
• 8.
  Architecture RASPBERRY PI AMAZONS3 BUCKET AMAZON SNS AMAZON SQS LAMBDA FUNCTION SMART HOME HUB
• 20.
  Result Expected : FireDetected but below threshold Actual : Fire Detected above threshold Expected : Fire Detected above threshold Actual : Fire Detected above threshold Expected : Fire Detected but below threshold Actual : Fire Detected but below threshold Expected : No Fire Detected Actual : No Fire Detected Following testing on different images and videos, we came to conclusion that we could reduce the false positive alarms to greater extent(35 – 40%).
• 21.
  Result Images (Input) Fire/ NoFire Images (Output) - Fire/ No Fire
• 22.
  Future Notes • Moreefficient light-weight model is needed on the pi for more efficient fire detection • By integrating thermal imaging with machine learning approach, we can further reduce false positive alarms • We can use buzzers, connected to pi which will ring once the fire is detected
• 23.
  Conclusion WE SHOWED THATFIRE ALERT SYSTEM CAN BE IMPLEMENTED USING IOT APPROACH WE SUCCESSFULLY, INTEGRATED AWS CLOUD SERVICES IN OUR SYSTEM WE SUCCESSFULLY, INTEGRATED DEEP LEARNING APPROACH FOR FIRE DETECTION
• 24.
  References 1. FireNet: ASpecialized Lightweight Fire & Smoke Detection Model for Real-Time IoT Applications (https://arxiv.org/pdf/1905.11922.pdf) 2. FireNet is an artificial intelligence project for real-time fire detection. 3. (https://github.com/OlafenwaMoses/FireNET) 4. Fire Now, Fire Later: Alarm-Based Systems for Prescriptive Process Monitoring(https://arxiv.org/pdf/1905.09568.pdf) 5. Insurance Information Institute (https://www.iii.org/fact-statistic/facts- statistics-fire) 6. http://hpthermalcamera.com/advantages-disadvantages-infrared- thermal-imaging-camera/
• 25.
  Thank you! Any Questions?