Inspiration
We realized that the Shading devices that are being used today to lower the temperature of the houses are controlled manually by the users. We wanted to come up with an AI algorithm that could automate the process, hence reducing the overhead drastically along with the cost.
What it does
This project builds two Deep Learning Architectures which uses just RGB images to control the Shading Devices automatically.
How we built it
We used one DL architecture to convert the RGB image into Thermal Image to detect the hot regions and another one to detect the glass windows. The two outputs are put together to find out the shading device that needs to be closed.
Challenges we ran into
Getting a proper dataset. Training a huge model like CycleGAN on a dataset like the RGB-T dataset which is 16 GB. Building a full fledged app with 24 hours.
Accomplishments that we're proud of
Coming up with a novel architecture which has not yet been used till now. Training the cycleGAN even though it was time consuming. Wrapping our architecture in a simple mobile application.
What we learned
How to convert RGB images to Thermal Images? The cost constraints of buying a thermal sensor. Practical difficulty of automating the shading devices and how to obviate them.
What's next for Ther-m-onitor
We wish to create a full fledged application for the user and also integrate our deep learning architecture to our IoT motor which can open and close based on the input it gets from the system. We also would like to collect relevant data for both the Deep Learning architectures and train them for longer durations, so that they output better quality images.
Log in or sign up for Devpost to join the conversation.