reacTIVision is an open-source, cross-platform computer vision framework designed for high-speed tracking of physical markers—known as fiducial markers—and multi-touch finger gestures. Created by Martin Kaltenbrunner and Ross Bencina, it was initially developed to power the reacTable, an interactive tangible musical instrument.
Instead of relying on heavy machine learning or standard square 2D matrix barcodes (like QR codes or ArUco), reacTIVision revolutionized fiducial tracking by utilizing a unique, math-driven topological approach. Core Innovations of reacTIVision
reacTIVision stood out from traditional computer vision systems by shifting the paradigm of how cameras interpret shape, structure, and communication.
[Camera Feed] ➔ [Adaptive Thresholding] ➔ [Region Adjacency Graph] ➔ [Tree Sequence Match] ➔ [TUIO Protocol Out] 1. Topology-Based Recognition (The Amoeba Marker)
Traditional markers rely heavily on rigid geometry or crisp squares. reacTIVision introduced organic, blob-like symbols frequently called “amoeba” markers.
Region Adjacency Graphs (RAG): The software converts a camera frame into a high-contrast binary (black and white) image using an efficient adaptive thresholding scheme. It maps out how black and white regions nest inside one another, building a hierarchical tree of concentric shapes.
Left-Heavy Depth Sequences: The system scans this tree using an algorithm called “left-heavy depth sequences” to match the structural tree against a preset dictionary.
Aesthetic and Functional Flexibility: Because tracking depends on nested topology rather than sharp angles, the markers still scan accurately even if the edges are distorted, blurred, or partially deformed. 2. Designed by Evolutionary Computation
The physical symbols weren’t drawn by hand. The creators used a genetic algorithm to “evolve” the amoeba markers.
mkalten/reacTIVision: computer vision framework for … – GitHub
Leave a Reply