Capturing what already happens in the room
Sensing gives the environment a readable model of what already happens in the room: entry, dwell, direction, crowding and change over time.
Refractiv defines which spatial events matter and how they affect the experience. The response can remain atmospheric, develop gradually or trigger a distinct state, according to the narrative and character of the project.
From raw input to usable behaviour
Most real systems combine detection with tracking. The detector proposes where people are in the frame; the tracker holds continuity over time, rejects noise and lets the installation work with dwell, direction, occupancy, grouping or zone crossing.
The media layer rarely needs pixels or bounding boxes. It needs clean behavioural events: someone entered a zone, two bodies merged into a group, dwell passed a threshold or occupancy changed. Depending on the brief, the pipeline may include segmentation, background modelling, world-coordinate mapping and event aggregation.
Designed around the actual scene
Public spaces bring changing light, reflections, crowds and varied movement. The sensing strategy is developed around the real conditions, with sensor choice, lens position, filtering, calibration and behavioural rules shaped through testing and site tuning.
Some briefs benefit from depth sensing or LiDAR when contrast is poor or geometry matters more than appearance. Others need a hybrid stack, such as camera-based detection with depth-assisted zoning, or overhead tracking combined with side-view confirmation.
Privacy-sensitive signal design
Where privacy requirements are strict, the system can be designed around anonymous spatial events, transient track IDs, volatile memory and event-level outputs.
The output is a sensing strategy that turns movement into usable behaviour while keeping the media layer stable, legible and appropriate for the space.
Technical signals
Inputs: movement, presence, proximity, dwell, direction, grouping, and occupancy.
Behaviour: detection combined with tracking, noise rejection, world-coordinate mapping, and clean behavioural events.
Media output: behavioural signals mapped to image, light, sound, or show control.
Control: computer vision, LiDAR and depth sensing, camera-based tracking, and hybrid stacks integrated in TouchDesigner.
Operation: sensor choice, lens position, filtering, calibration, and site tuning, with privacy-sensitive event design where required.
Where it is used
Retail, public art, sport and brand activations, and any space where movement should drive stable, legible media behaviour.
For a commission or an early-stage sensing enquiry, get in touch.
