InfoMus Lab
The EyesWeb Motion Analysis Library
The EyesWeb Motion Analysis Library (some parts can be downloaded here) applies computer vision, statistical, and signal processing techniques to extract expressive motion features (expressive cues) from human full-body movement.
At the level of processing of incoming visual inputs the library provides modules including:
-
Background subtraction techniques for segmenting the body silhouette
-
Techniques for individuating and tracking motion in the images from one or more videocameras
-
Algorithms based on searching for body centroids and on optical flow based techniques (e.g., the Lucas and Kanade tracking algorithm
-
Algorithms for segmenting the body silhouette in sub-regions using spatio-temporal projection patterns
-
Modules for extracting silhouette's contour and computing its convex hull
|
|
|
| On the left: contour and convex hull of the body silhouette; on the right: sub-regions of the body silhouette and body center of gravity. | |
At the level of extraction of motion features a collection of parameters is available. Some of them are listed below:
-
Quantity of Motion, i.e., amount of detected movement
-
Contraction Index
-
Stability Index
-
Asymmetry Index
-
Silhouette shape and direction of body parts
|
|
| Silhouette shape and orientation. The ellipse approximates the silhouette; its axes give an approximation of the silhouette’s orientation |
The EyesWeb Motion Analysis Library also includes blocks and patches extracting measures related to the temporal dynamics of movement. A main issue is the segmentation of movement in pause and motion phases. Several movement cues can be measured after segmenting motion in motion and pause phases: for example, blocks are available for calculating durations of pause and motion phases and inter-onset intervals as the time interval between the beginning of two subsequent motion phases.