An intelligent wildfire prevention platform using an AI-powered camera and computer vision
UX/UI Designer and Pitch Presention
Jhorman Perez & Andrés Jiménez Pacheco
September 24-26, 2021
September 2018 - July 2019
Stata Mater was created as a submission for the 3-day Ultrahack's Health and Safety Monitoring DeepHack in 2021. The challenge was to create a data-driven solution that helps individuals and society to stay safe and healthy using edge AI devices. Since this hackathon was sponsored by Luxonis, each hackathon participant recieved a Luxonis LUX-ESP32 spatial AI camera to be utilized for the solution. The hackathon was also online, so my team was divided between New York City and Berlin.
As a result, we came up with Stata Mater, an easy-to-use platform that uses artificial intelligence and computer vision to help people with wildfire prevention through a reward system approach. By identifying risks and practicing good prevention techniques, users can get rewarded while helping a global issue.
Our teamwork won 3rd Place Overall and was praised on its design, object detection model, and pitch presentation by the judges, sponsors, and mentors.
Andrés and Jhorman already had an ongoing idea for wildfire prevention through real-time risk assessment and artificial intelligence, but it was at a very early state. We started off at this point for brainstorming and did some generative research for submission of the hackathon. We knew that there were many solutions for predicting wildfires and analyzing live wildfires, but there was a lack of solutions for wildfire prevention. Therefore, be wanted to tackle risk management so that we could prevent the issue of wildfire happening in the first place.
Since our team was split between Germany (Andrés and Jhorman) and the United States (myself), we had some video calls and used Miro to do some virtual brainstorming sessions. We came up with ideas on post-it notes that our solution could provide and used dot voting to synthesize our project.
After further discussion and research, we boiled down our project to the following:
A AI-powered platform for wildfire prevention by identifying property risks with a gamified approach on insurance
To combat wildfires, we introduce Stata Mater: a mobile app that allows property owners to quickly and seamlessly identify high-risk objects for wildfire and to improve the overall experience for everyone. In Ancient Roman Religion, Stata Mater is the name of a goddess who protected against fires. We plan to do the same with our platform by moving insurance from a traditional claims-payout solution to one that’s much more fun and preventive.
Stata Mater provides technology that crushes the growing threats of wildfire along with protecting owners and their properties. Stata Mater makes the insurance process easier by creating motivation for users to have a safer home against wildfires.
In just a few taps and a working AI camera, Stata Mater helps user get started and avoids the tedious process of dealing with insurance in this one hub. Our design is made to be user-friendly, easy, and stress-free.
With our AI-powered camera and technology, Stata Mater makes it so much easier for users to find threats like dry vegetation and flammable materials that can potentially contribute to wildfires. Also, the conversational UI makes it friendlier for users and delivers a personalized user experience.
Users can collect points to redeem for discounted insurance and other perks. It's a win-win situation for all involved: this method attracts customers, reduces claims, keeps people safe, and also helps the environment.
In the end, we were able to have a running object detection model that was able to identify dead vegetation, flammable property materials, and fire-prone mulches along with an interactive app prototype.
Our workflow for the development of our prototype was as follows:
1. YOLOv4 by @AlexeyAB on Github was used at the base object detection architecture for a good edge processing on the portable device, and it has improvements in terms of image processing to make detections more accurate. We chose this because it had reliable performance for onboard processing, and has improved data management and data augmentation in the training pipeline.
2. To build the dataset, we used an automatic search engine to pull images from Google and Bing. We also used an automatic bulk image download by @tomahim on Github for some assistance. Then we annotated them with labels of dry vegetation, flammable property materials, and fire-prone mulches. and apply image augmentation techniques, for a bigger dataset. Sometimes additional image augmentation was applied by flipping the image and considering exposure, brightness, and blurriness.
3. Finally, to get our AI ready to go into production, we had to convert YOLO to Tensorflow to make it supported by OpenVINO. Then, it was just about the intermediate representation and optimization, and then to put it into the Luxonis device.
Stay positive. During a hackathon, it's easy to feel like things aren't coming together due to stress and time constraints. When everyone is working so hard and lacking on sleep, sometimes our mood can get the best of us. You can usually see other people's submissions and progress at hackathons, and so it's easy to compare yourself to the other amazing work that teams are doing. However, doing this takes away energy from yourself and self-esteem, which can be deterimental in the end. Therefore, I learned that it is important to stay calm and believe in my team.
Make decisions. At first, so many ideas and features were thought to be included in the final prototype. However, with such limited time at a hackathon, it's important to realize that having too many things to include could really spread out the team's energy and effort. Time management is incredibly key in a hackathon! Thus, we made it a point to make decisions on what is feasible and relevant, and to focus on quality rather than quantity.
Be proactive. It is especially important to be communicative with each other when our team was split up among continents and time zones due to the nature of an online hackathon. We had to be comfortable with each other to never hesitate to ask questions and check up on each other constantly since we weren't physically together.