NULL Pointers| Spot That Fire V3.0
Awards & Nominations
NULL Pointers has received the following awards and nominations. Way to go!
Spot That Fire V3.0
Volca: The Fire Finder
Summary
Our idea is based on three major entities; Analyse - Reconnaissance - Mitigate. The implementation includes having mechanisms to -Aggregate wildfire information from MODIS and VIIRS satellites and via crowd-sourcing by allowing users to report wildfires with a smartphone.Validate and map wildfires via autonomous drones that are capable of recognizing fires utilizing Computer Vision. Drones are more effective in evaluating the rate of spread as they are immune to smoke and thick forest canopies. Using this, we predict the extent of rate and direction of spread. Drones also help to find stranded humans and wildlife and alert the rescue authorities.
How We Addressed This Challenge
Our idea is based on three major entities; Analyse - Reconnaissance - Mitigate. The implementation includes having mechanisms to -
- Aggregate wildfire information from MODIS and VIIRS satellites and via crowd-sourcing by allowing users to report wildfires with a smartphone.
- Validate and map wildfires via autonomous drones that are capable of recognizing fires utilzing Computer Vision. Drones are more effective in evaluating the rate of spread as they are immune to smoke and thick forest canopies. Using this, we predict the extent of rate and direction of spread. Drones also help to find stranded humans and wildlife and alert the rescue authorities.
- Help in mitigation by providing aggregated data from drone reconnaissance to fire-fighters to help curb the spread and to individuals to plan their escape routes.
It is important to quickly reduce the detection and mitigation turn-around times. With this, we should be able to -
- pin-point high confidence geo-thermal anomalies
- verify them with drones and crowd-sourcing
- help predict the extent of damage
- find trapped humans and animals
- provide escape routes
How We Developed This Project
Development was a monumental task with the aggregation of various technologies.
We needed to build an object detector to detect fire, a crowd-source application, a back-end API service to serve the object detector and a drone control system.
The Neural Network -
We trained CenterNet-ResDCN34 object detector to recognize people and fire. The person data was extracted from the infamous coco-dataset and for fire, we web-scrapped 5000 images of wildfires from google. We used Pytorch with GPU acceleration and served it using Flask.
The drone -
We used a DJI Ryze Tello drone, which has an open-source sdk. We interfaced with it using TelloPy and Python. Using it's onboard camera, we could stream the data and use our object detector to recognise fire.
The crowd-source application -
Our crowd-source application is built with flutter.
Our functionality included:
- Reporting Fires with Geo-locations that are validated by using the aforementioned neural net
- Providing areas of confirmed Wildfires and Escape Routes
- General instructions to individual in the event of a fire
How We Used Space Agency Data in This Project
Our main source of data was NASA's Active Fire data
https://firms.modaps.eosdis.nasa.gov/active_fire/#firms-txt
We used the CSV files to aggregate the known locations of plausible thermal anomalies.
It consisted of lat-long coordinates and confidence values. We set a threshold and filtered/clustered the values.
Project Demo
Find the presentation at : https://prezi.com/p/nyutxfbgpl6v/?present=1
Find the code at: https://github.com/aj-ames/NSAC-NULLPointers
Data & Resources
- https://firms.modaps.eosdis.nasa.gov/active_fire/#firms-txt
- https://github.com/xingyizhou/CenterNet