
This tool is designed as a real-time multi-tracking system for interactive games and installations using video input from surveillance cameras. It identifies pedestrians, tracks their position and size, detects specific actions, and sends this filtered data via the OSC protocol (other network protocols can also be integrated).
Built with the YOLOv2 object detection model using the Darknet library, it processes video on the GPU, allowing real-time performance — tested at around 17 FPS on a Nvidia GTX 1060.
The system performs reliably even in low-light environments, making it suitable for urban settings or nighttime installations.

→ More info about Medialab-Prado’s Digital Facade
Get the App
If you’re a client or user, you can download executable versions and Processing client examples from the Medialab-Prado Github repository, under the sensor4Games/yolo2
folders.
For Developers
Check out the source code and documentation on the main GitHub repository.
Requirements
- OS: Windows 10
- GPU: Nvidia Pascal series (tested on GTX 1060)
- Dependencies: Darknet, OpenCV, OSC support (e.g. ofxOSC for Processing or liblo for C++)