This ROS package is designed for early wildfire detection, geolocation and monitoring.
| functions | results |
|---|---|
| path planning |  |
| forest fire image classification |  |
| forest fire image segmentation |  |
| gimbal control |  |
| fire point geolocation |  |
Early wildfire spot perception methods
| functions | |
|---|---|
| Attention gate U-net wildfire segmentation |  |
| Trianglulation-based wildfire point depth estimation |  |
| Visible-infrared camera system calibration |  |
| Model-based wide fire point registration |  |
You can also compare the reconstruction results with google earth.
SFM with colmap |
Reconstruction with OpenMVS |
|---|---|
 |
 |
 |
 |
M300 drone flies along the defined zigzag path, while the attention gate U-net is implemented to detect and segment the suspect wildfire.
- Once the suspect fire is detected, M300 will fly from left to the right. In the meantime, a monocular SLAM is running to acquire the precise camera pose, the GPS information is used to correct the scale of the poses.
- Based-on the triangulation of the fire zone, the distance can be estimated. Then, the fire can be geolocated by with the gimbal angle.
Finally, M300 will fly along a circle shape of the flight path to record the scene of the local environment around the wildfire spot with the H20T zoom camera.
@INPROCEEDINGS{9836119,
author={Li, Shun and Qiao, Linhan and Zhang, Youmin and Yan, Jun},
booktitle={2022 International Conference on Unmanned Aircraft Systems (ICUAS)},
title={An Early Forest Fire Detection System Based on DJI M300 Drone and H20T Camera},
year={2022},
volume={},
number={},
pages={932-937},
doi={10.1109/ICUAS54217.2022.9836119}}Copyright (C) 2021 Concordia NAVlab. All rights reserved.