Wildfire Asset Tracking via Sensor Fusion and Machine Learning

Problem Statement

During wildfires, hundreds to potentially thousands of firefighters and scores of firefighting equipment are deployed to contain the fire. It is critical to accurately track such deployed assets to coordinate resources, protect firefighters, and enhance situational awareness. Currently, most assets are tracked through radios or rudimentary techniques, such as visual light communication, which often results in poor situational awareness in active fire situations. RFID technology has been utilized in a wide array of tracking applications both indoors and outdoors, and has been shown to achieve high accuracy. However, there has been limited exploration of its use in wildfire applications.

Approach

This project proposes the incorporation of RFID technology into wildfire asset tracking through the use of a mobile autonomous robots, leveraging Gaussian Processes for environmental analysis and range prediction. This research address challenges in environmental selection, range prediction, camera RFID association, and persistent monitoring of mobile assets. The key contributions of this research include a novel application of a Gaussian Process model, the ability to identify environments based off of RF signal without any spatial information, fusion of camera and RFID tags under challenging conditions, and algorithms to persistent monitor mobile assets that move in and out of detection range. This study aims to add a new application of RFID technology and enhance wildfire situational awareness through the incorporation of a low cost tracking system.

Figure 1 – Overall vision for the camera-RFID system where the sensors and RFID reader(s) are mounted on a mobile platform and used to track various human and non-human assets during firefighting situations.

 

Ongoing work

The current focus of this work is fusing LIO-SLAM with obstacle avoidance schemes and custom persistent monitoring techniques to track mobile assets as they move through the area and outside the detection range of the RFID system.

Key Results and Visuals

Four participants moving randomly in the test environment while each holding an RFID tag. Bounding box detections tracking individual participants also shown.

 

Relevant Publications

  • Tracking Wildfire Assets with Commodity RFID and Gaussian Process Modeling (IEEE RFID, under revision)
  • Camera-RFID Fusion for Localization in Forested Environments (ICRA 2026: Submitted)

 

Students

John Hateley