Ultimate Occupancy

PIR and Grid-Eye based occupancy sensing

source: Wired
Machine Learning
Signal Processing
Prof. Mayank Goel
Dhoyun Kim
Jordan Stapinksy

This project is a purely technical exploration that involved working with two other software engineers, Dhoyun and Jordan, to create a sensing system that could detect the occupancy level of a room without a camera or any recorded video. 

Instead, we used CMU’s Super Sensor (pictured above) to collect passive infared sensor as well as 64 channel IR data from the onboard ‘grid eye’ sensor in order to do this. Even though this data is very noisy, we developed a machine learning model that could be trained to read these sensor values and make inferences about the number of people in that room. 

My role on the project was actually not in a design capability in almost any way. I trained and built the machine learning module of this project. I was responsible for all the initial experimentation all the way to implementing a model in our final application. Jordan handled feature extraction from our sensor data and Dhoyun handled the code that actually ran on the sensor as well as built the main infrastructure for the project. The supersensor hardware itself was developed by the Future Interfaces Group here at CMU. 

See my other (very rough!!) signal processing/iot sensing experiments HERE.

Further technical details in our paper below: