Depth cameras — Microsoft’s Kinect line chief among them — remain one of the fastest ways to get real-time body position and movement data into an interactive installation without writing custom computer-vision code from scratch. Paired with TouchDesigner, a node-based visual programming environment built for exactly this kind of real-time, sensor-driven work, a depth camera becomes the input side of an interaction loop you can build in an afternoon.
Watch: Kinect Azure Point Cloud in TouchDesigner - TouchDesigner Tutorial 068 (The Interactive & Immersive HQ, YouTube)
Step 1: Know which Kinect you actually have
Kinect Azure (Microsoft’s more recent depth sensor, technically the “Azure Kinect DK”) and the older Kinect v2 both work with TouchDesigner, but through different operators and with meaningfully different capabilities — Azure offers better depth resolution and a wider range of capture modes. Confirm which hardware generation you’re working with before following a tutorial written for the other; the setup steps and available parameters differ enough to cause real confusion if you’re troubleshooting the wrong operator’s documentation.
Step 2: Get a point cloud on screen first
In TouchDesigner, the Kinect Azure TOP operator handles the camera connection directly — no separate driver installation dance beyond Microsoft’s official Azure Kinect SDK. The fastest possible confirmation that everything is wired up correctly is producing a basic colored point cloud: a 3D scatter of points representing everything the depth sensor can see, colored using the camera’s RGB feed. This requires only a handful of operators and gives you immediate visual proof the hardware, drivers, and software are all talking to each other before you build any actual interaction logic on top.
Step 3: Turn depth data into interaction, not just visualization
A point cloud is a visualization — the interactive part comes from deciding what triggers what. Common first patterns: mapping a visitor’s proximity to the sensor to a parameter (brightness, particle density, sound volume), using optical flow (motion between frames) to drive particle systems that react to how fast and where someone is moving, or defining a threshold zone in 3D space that triggers a discrete event when crossed. Start with proximity-to-parameter mapping specifically — it’s the simplest possible interaction to get feeling right, and most of what makes an installation feel responsive versus laggy comes down to tuning that one relationship before adding anything more complex.
Step 4: Design for a room, not a desk
The single biggest gap between a working desktop test and a working installation is environment: depth sensors are sensitive to ambient infrared light (direct sunlight is a common failure point), sensor placement height changes what the camera actually captures of a person’s body, and multiple simultaneous visitors can confuse simple threshold-based logic that was only tested with one person in frame. Test in the actual install environment, at the actual mounting height, with more than one person in frame, well before a show date — these are the failure modes that don’t show up in a quiet studio test.