Our team members are keen on computer vision and deep learning. Helvar’s challenge with intelligent light switch has inspired us to build a project at the junction of state of the art computer vision and people's everyday lives.
What it does
Life moves too fast, bombarding us with too much information, tempting us with too many distractions from what really matters. That’s why it is so essential to make an effort once to make our lives easier forever!
Think of you lying on the bed feeling like falling asleep but you have to get up to turn the lights off. Bruh. Just imagine that you don’t have to do it anymore! All you need to do is just point your hand to the light bulb and voila!
People are too lazy to control the lights manually! We want to help them by automating the whole process and giving the user a really frictionless control. After all, let’s be honest. What can be cooler than a possibility to “rule” the lighting with an elegant hand movement just like Albus Percival Wulfric Brian Dumbledore did?
How wonderful it would be if we could add some magic to our non-magic lives!
What we offer is controling lightning system effortlessly just with hand pointing up at a light bulb! For the solution we use camera and artificial intelligence to analyze gestures and REST API to manage the lights.
How does it work
As simple as you can imagine: when a person points to light bulb it just inverts its state - turns on, or turns off accordingly.
Video demonstration 1: https://youtu.be/snDutNGH348
Video demonstration 2: https://youtu.be/EJGipiplMSM
Python application which:
- Connect to camera and read frames using OpenCV.
- Send HTTP requests with needed frames to the server with pose estimator.
- Switch the light accordingly to server's responses using Helvar REST API.
Python Docker Service which:
- Get HTTP requests with frames.
- Estimate pose of persons on them using PyTorch machine learning model. (State of the art paper “Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields”)
- Determines if some hand directed to the bulb using algorithm based on vector algebra.
- Send HTTP response to the client.
Challenges we ran into
We faced with challenge to match predicted pose with light we want to control, after few experiments we decided to use a line between a shoulder and an arm, computing the length of the shoudler-elbow and shoulder-arm we can understand that a person wants to switch the light on/off.
What's next for HandyLight
As technology achieves great progress in machine learning, the idea of smart homes is starting to materialize. Nowadays there is a lot of amazing project for making our home easy and cozy. Obviously, the light is essential part of every home. That’s why our idea is to make light not usual process of turning on/off the lights, but simplified version of it wich works with hand motion.
We hope that during the judging we will get inspiring suggestions or feedback on our project.