Inspiration
Based on the recent devastating fire events around the globe, a quick to respond and autonomous solution needed to be made to notify the authorities and nearby residents to evacuate or take immediate action.
What it does
It autonomously patrols a predefined area providing a live feed to the authorities. Using deep learning algorithms and a plethora of model data to be trained, the model accurately detects fires and notifies the authorities.
How we built it
Using our available hardware and our trained model built using python and utilising Google's TensorFlow along with the necessary Twitter and Twillio APIs to notify, our finished solution patrols the area and detects fires using the visual sensor based on the model data it was trained on.
Challenges we ran into
Live streaming the drone's video stream and live processing it without a significant lag was a challenge, as well as combining all the elements together in controlling the drone and accurately detecting fire amongst other elements in the input images.
Accomplishments that we're proud of
The final solution is incredibly accurate, relatively responsive and reliable in notifying the authorities when it detects fire.
What we learned
Deep learning can be incredibly difficult to fine tune to accomplish what you want. Live processing continuously streaming data and providing a useful output can also be quite challenging, but we figured it out in the end.
What's next for UAV FireWatch
Expanding further than our hardware limitations, most importantly a thermal camera can be utilised, thermal coating on the drone to be able to better approach fires and send information and a more accurate location info system for accurate route planning and location info on fire events would be the next immediate step in our project.
Log in or sign up for Devpost to join the conversation.