Forest Fire Tracking from Aerial Images and Measuring its Severity/Intensity.
- Mentors
- Arghya, Yali Wang
- Organization
- Alaska
- Technologies
- python, mongodb, react, opencv, tensorflow, MLflow
- Topics
- computer vision, deep learning, Object Detection, System Design
Forest fires in the boreal regions, particularly in Alaska, result in significant losses, ranging from loss of life to environmental degradation and disruption of essential services. Detecting and tracking these fires swiftly and accurately is crucial for effective rescue operations and resilience planning. This project seeks to develop precise metrics for assessing the severity and intensity of forest fires, enabling better management strategies.
Introducing a novel approach where video frames are simultaneously passed through two methods: the first method focuses on detecting fire spots and accurately estimating the area of fire spread using bounding boxes and fire pixel analysis. The second method analyzes the direction and speed of the fire spread by calculating optical flow vectors.
Forest rangers will receive actionable insights in real-time, including the area, direction and speed of fire spread, along with the count and location of fire spots, the distance between spots, and alarm functionalities to facilitate quick decision-making. These insights will be geo-referenced using drone flight data, providing accurate coordinates for targeted intervention and containment efforts.
Set of Deliverables:
- An optimized fire detection model capable of accurately identifying fire spots.
- Parameters and algorithms for calculating severity based on fire detection data.
- A comprehensive system for providing real-time actionable insights to forest rangers, enabling them to make informed decisions and respond promptly to forest fire incidents.