← All research

01 Reveal hidden traces

Body weight estimation

We estimate human body weight from music-induced bed vibrations. The main intuition is that body weight changes how vibrations propagate through the bed–body system, and these changes are captured by the transfer function measured between two vibration sensors.

We design music to excite weight-sensitive frequency bands and estimate weight with a neural network whose activation functions are informed by the theoretically derived weight–vibration relationship.

See how it works Body weight estimation · Preprint 2025 ↗

How weight changes the bed–body transfer function

MEASURED DATA

Choose a recorded load to compare how music-induced vibrations propagate from the reference sensor beside the speaker to a sensor across the bed.

The animation shows music exciting the bed and vibrations reaching the sensors.

Drag horizontally on the person, use the arrow keys, or use the recorded-load slider below.

Drag the person or adjust the load below.

Measured transfer-function magnitudes over 100–1000 Hz.
53.7055.0656.0657.06 kg

Four measured conditions · no values interpolated between loads

Loading measured transfer functions…

Selected load53.70 kg referenceMiddle 50% of windows

Falling into Shadows

1:40

Music from the experiment.

From transfer function to body weight

Scroll to explore the figureView full figure
A speaker on the left and response vibration sensor S3 on the right sit symmetrically at mid-body height. Reference sensor S1 is beside the speaker. The measured transfer-function magnitude from S1 to S3 enters a neural network guided by bed–body structural dynamics.

Through structural-dynamics analysis, we derive how body weight affects the bed–body transfer function. A rational approximation of this relationship informs a learnable Padé activation, whose coefficients account for unknown bed properties. Paper ↗

Method & references

The transfer-function branch adopts a Padé Activation Unit (PAU), matching the rational approximation derived from the bed–body vibration model. Its trainable coefficients represent structural influences such as stiffness and boundary conditions. The height branch uses a learnable sine activation motivated by structural mode shapes, with height indicating bed–body contact area. The two branches feed the weight regressor.

PAPERS

  1. [1]

    Y. Wu, J. Zhang, M. Lee, C. Smith, X. Li, A. Senapati, P. Zhang & H. Y. Noh (2025). Human Body Weight Estimation Through Music-Induced Bed Vibrations. arXiv preprint.Theoretical derivation: §2. Physics-informed activation functions: §3.2.3.

  2. [2]

    Y. Wu & H. Y. Noh (2026). Non-contact Mass Estimation of Static Objects on Kirchhoff–Love Plates via Active Vibration Sensing. SSRN preprint.

← Explore the other projects