Clarkson University recently installed a new makerspace to facilitate student, faculty, and staff projects. Given that it was a new space, and various decisions about organization, open hours, and equipment purchasing had to be made, we sought to provide administrators with quantitative data about utilization patterns. The system provides a heatmap visualization demonstrating the frequency of utilization over a period of time. This will be presented in an easy-to-use website, to make the data easily accessible. This work won the second Clarkson University President's Challenge, for a grand prize of $7,000. Additional content from the contest can be found here: Slides, Video, Paper, Demo Website.
We process data from a mounted camera on an Nvidia Jetson, using OpenPose to extract people's locations, and use a homography to map this data into an overhead view. From this, we generate intuitive visualizations on how the space is used over a specific interval.
We are currently working on setting this up so it runs in real time and publishes to a publicly-available website. Additionally, we would like to look at capturing skeletons in 3D, rather than 2D, likely with the addition of multiple cameras. We welcome contributions.
David Russell, Damon Gwinn, Nikolas Lamb, and Adam Romlein.
Run the ./example.sh script
To install openpose on Jetson:
- Flash with jetpack 4.2.2
- Ensure Jetson is in max power mode
sudo nvpmodel -m 0. - Install openpose dependencies following their tutorial here:
https://github.com/CMU-Perceptual-Computing-Lab/openpose. - Checkout this openpose commit
git reset --hard 06d4ea6(this is to avoid this error: CMU-Perceptual-Computing-Lab/openpose#1392) - Upgrade cmake to 1.15.4 by building from source.
- Create build directory, cd into it, run
cmake -DBUILD_PYTHON:=ON ... - Run make -j6.
- Test python installation by following these instructions:
https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/modules/python_module.md.