Zuchen Li

All projects

MobiCom 2026 DemoFirst Place, ACM SRC (Undergraduate) Ubiquitous sensing

MoiréDeform

Towards Fine-Grained Deformation Sensing through Naturally Occurring Moiré Patterns in Everyday Woven Surfaces

Linzhen Zhu*, Zuchen Li*, Weihao Jin, Hyunmin Park, Alanson Sample, Ke Sun

University of Michigan — * equal contribution

Co-first author, Human Centered Computing Lab, University of Michigan, with Profs. Alanson Sample and Ke Sun

MoiréDeform: a clip-on camera observes an unmodified mesh chair back (the user is never in frame); subtracting an empty-chair reference yields the moiré pattern-difference view, which drives live respiratory-waveform reconstruction with breath-hold detection and body-lean joystick interaction.

Every time we sit, lean, or breathe, we gently deform the surfaces that hold us. Those tiny deformations carry signals about posture and even breathing — but they are hard to see without instrumenting the surface or the person.

Moiré patterns make them visible. When two fine, repeating patterns overlap, a tiny shift between them turns into a large, sweeping fringe. Woven surfaces like the mesh of an office chair already carry that repeating pattern. MoiréDeform turns the mesh into a sensor: a low-cost camera behind the chair records one reference frame of the empty backrest, and each live frame is compared against it, so the two act as virtual layers and deformation appears as a moiré pattern. No markers, no added gratings, no changes to the chair — and the person sitting in it is never in frame.

What it does

  • Breathing. Each breath presses gently into the mesh. A lightweight neural network reconstructs the breathing waveform live from the moiré signal, and flags a breath hold within about 1.5 s.
  • Posture as input. Shifting your weight across the backrest moves the load on the mesh; the weighted centroid of the moiré envelope becomes a continuous two-dimensional joystick — here steering a seated back-mobility exercise.

Results

In a preliminary study with 5 users and 100 minutes of recordings, MoiréDeform tracked breathing rate within 0.72 ± 0.45 breaths/min of a chest belt. With the same camera, the moiré signal was about 3× more accurate than tracking a physical marker (0.71 vs. 2.07 breaths/min error). The same channel could support stretch-break coaching, posture awareness, hands-free input, or games you play by leaning — all by watching fabric, not faces.

Videos

  1. 30-second teaser 0:30
  2. Narrated demo 2:45

Figures

A tiny shift between two fine patterns becomes a large, sweeping moiré fringe.
Figure 1. A tiny shift between two fine patterns becomes a large, sweeping moiré fringe.
The chair's own weave provides the repeating pattern — no markers or gratings added.
Figure 2. The chair's own weave provides the repeating pattern — no markers or gratings added.
In use: the camera sees only the back of the chair, never the person.
Figure 3. In use: the camera sees only the back of the chair, never the person.
Demo 1: the breathing waveform reconstructed live by a lightweight CNN.
Figure 4. Demo 1: the breathing waveform reconstructed live by a lightweight CNN.
A breath hold is flagged within about 1.5 seconds.
Figure 5. A breath hold is flagged within about 1.5 seconds.
Demo 2: the weighted centroid of the moiré envelope becomes a 2D joystick.
Figure 6. Demo 2: the weighted centroid of the moiré envelope becomes a 2D joystick.
A seated back-mobility exercise, steered by posture alone.
Figure 7. A seated back-mobility exercise, steered by posture alone.
Preliminary evaluation: 5 users, 10 sessions, 100 minutes.
Figure 8. Preliminary evaluation: 5 users, 10 sessions, 100 minutes.