Media Technology MSc

Bartertown: Dataset of Iconic Gestures

This page hosts a dataset of iconic gestures by human participants, depicting shapes. The dataset is compiled by non-experts through a human-computation game (a.k.a. game with a purpose, or gamified crowd-sourcing). Researchers interested in gesture-recognition and gesture-generation can use this data for their studies.

This work was part of the graduation project of student Wouter van den Heuvel (2015). More information can be found in Wouter's thesis.

Game

The eponymous game behind the dataset was created using the Unity 4.6 game engine and uses a Microsoft Kinect for Windows v2 for interaction and gesture capturing. The game was tested during an experiment in April-May 2015 with 36 voluntary participants. They each performed four gestures referring to different shapes, after which they had to validate existing gesture performances (demonstrated through a virtual agent). During the validation, participants could choose between four different shapes (one of which was the actual referent shape, i.e. ground truth) or choose "I don't know". The video below briefly demonstrates the test environment.

Shapes vocabulary

These are the shapes that form the gesture vocabulary. They are considered iconic gestures and refer to primitive shapes.

iconic shapes dataset

Dataset

The dataset contains tracked skeleton and labelling data of all participants. The skeleton data is stored as positions and rotations of each Microsoft Kinect v2 recognised joint in 3D space at 30 fps.

The dataset can be downloaded in two formats, each with a readme file explaining the structure of the data:

  • CSV format; an export of the data in comma-separated files, divided into seperate folders.
  • SQLite format; a relational database using the  SQLite engine.

Using the data

The data can be freely used for scientific purposes. Although the utmost care was taken in collecting, preparing and exporting the data, no warranties regarding consequences of its use are given.
When using the data, use the following citation regarding its source:

Wouter van den Heuvel, "Bartertown: Dataset of Iconic
Gestures", Media Technology MSc program, Leiden University
(2015). www.mediatechnology.leiden.edu/openaccess/bartertown

Should you have any questions about the dataset, contact ....