Theoretical Framework

PyEyesWeb inherits from the rich tradition of computational movement analysis initiated by the EyesWeb project ¹²³ and grounds on the multi-layered computational framework of qualities in movement developed in the DANCE project ⁴.

Conceptual Model

The framework describes how raw sensor data can be progressively transformed into meaningful descriptions of expressive movement qualities and is organized into four layers.
The layers represent a conceptual model and not a strict processing pipeline.

Layers Overview

Attention!

The concept of timescale is crucial in this framework, and each layer operates at different temporal scales.
As an example, a key distinction from Layers 2 and 3 is moving from instantaneous or short-window features (~0.5s) to longer windows (0.5-3s) or movement units (e.g., a specific sport gesture, a choreographic phase).
One same feature can occur at different layers and yield different interpretations depending on the timescale of analysis.

• Layer 1 – Physical Signals

  Raw data captured by virtual sensors, i.e., physical devices (motion capture, accelerometers, video, RGB-D cameras, physiological sensors, etc.) enriched with preprocessing (denoising, filtering, extraction of trajectories, silhouettes, respiration, etc.).

  Foundation for all higher layers.

  → Learn more

• Layer 2 – Low-Level Features

  Instantaneous or short-window (0.5s) descriptors computed from physical signals.
  Includes: velocity, acceleration, kinetic energy, Quantity of Motion (QoM), postural contraction, balance, smoothness, etc.

  Represented as time-series.

  → Learn more

• Layer 3 – Mid-Level Features

  Operates on movement units or longer windows, producing structural descriptors in multidimensional spaces.
  Examples: contraction/expansion, symmetry, directness, lightness, suddenness, fluidity, repetitiveness, coordination, cohesion.

  Introduce amodal descriptors across modalities.

  → Learn more

• Layer 4 – Expressive Qualities

  Focuses on what an observer perceives from movement: emotional expression, social signals, saliency, attraction/repulsion, groove, hesitation, predictability.
  Involves memory and context, influencing how movement is interpreted (expectancy, contrast, sensitivity).

  Requires context and ML mappings.

  → Learn more

• Analysis Primitives
  Core computational tools applied across all layers.
  Includes: statistical moments, entropy, shape descriptors (peaks, slopes), synchronization, time-frequency transforms, predictive and physical models (e.g., mass–spring).

  Provide the building blocks for extracting meaningful features.

  → Learn more

References

---

1. Camurri, A., Mazzarino, B., & Volpe, G. (2003, April). Analysis of expressive gesture: The eyesweb expressive gesture processing library. In International gesture workshop (pp. 460-467). Berlin, Heidelberg: Springer Berlin Heidelberg. ↩

2. Camurri, A., Coletta, P., Massari, A., Mazzarino, B., Peri, M., Ricchetti, M., ... & Volpe, G. (2004, March). Toward real-time multimodal processing: EyesWeb 4.0. In Proc. AISB (pp. 22-26). ↩

3. Volpe, G., Alborno, P., Camurri, A., Coletta, P., Ghisio, S., Mancini, M., ... & Sagoleo, R. (2016). Designing multimodal interactive systems using EyesWeb XMI. In CEUR Workshop Proceedings (pp. 49-56). CEUR-WS. ↩

4. Camurri, A., Volpe, G., Piana, S., Mancini, M., Niewiadomski, R., Ferrari, N., & Canepa, C. (2016, July). The dancer in the eye: towards a multi-layered computational framework of qualities in movement. In Proceedings of the 3rd International Symposium on Movement and Computing (pp. 1-7). ↩