Designer and engineer interested in computational archaeology: recovering abandoned computing paradigms and rebuilding them with modern tools.
Also a wooden boatbuilder.
Former Recurser.
Currently ITP @ NYU
Designer and engineer interested in computational archaeology: recovering abandoned computing paradigms and rebuilding them with modern tools.
Also a wooden boatbuilder.
Former Recurser.
Currently ITP @ NYU
Satellite
A kinetic sculpture that uses primitive edge detection to navigate reconnaissance satellite imagery, autonomously panning to structures it identifies as human-made.
The Concept
Satellite uses satellite and reconnaissance imagery provided by the Defense Innovation Unit Experimental (DIUx) and the National Geospatial-Intelligence Agency (NGA) — datasets originally released to encourage development of overhead object-detection algorithms.
Instead of training a neural network, I used one of the oldest object-of-interest algorithms: simple edge detection. The machine scans these military reconnaissance images with the visual acuity of 1970s computer vision, hunting for human-made structures with methods that predate the surveillance apparatus that captured the images.
Where it decides to go, and in what order, is up to the algorithm.
The Installation
Physical form:
- Modified pen plotter mounted on gallery wall (36" × 36" × 7")
- XY gantry system with exposed stepper motors and drivers
- Small display mounted on the moving carriage showing current view
- Larger monitor displaying the full image with edge detection overlay
- Power supply, wiring, and electronics exposed
Behavior:
- The plotter slowly pans across satellite imagery
- Edge detection identifies potential structures
- The gantry physically moves to center on detected objects
- When a path completes, the system loads the next image
- The machine operates autonomously
Technical Details
- Vision: openFrameworks (C++) with threshold-based edge detection
- Motion: Custom plotter drivers, serial communication to stepper controllers
- Display: Dual screens — one on the moving gantry, one stationary reference
- Control: Mac Mini running the openFrameworks application
Why Primitive Algorithms?
Modern ML-based object detection would be more accurate, but that's beside the point. By using obsolete computer vision techniques to analyze state-of-the-art surveillance imagery, there's a temporal collision: 1970s perception applied to 2020s reconnaissance. The machine's limited understanding - its tendency to mistake shadows for structures, to fixate on roads, to miss obvious buildings - is the subject.
Exhibition History
Perimeter Group Show, All Street Gallery, 2025
Source Code
- Main application: [github.com/nikokozak/satellite](https://github.com/nikokozak/satellite)
- Hardware drivers: [github.com/nikokozak/satellite_drivers](https://github.com/nikokozak/satellite_drivers)