A motion capture library for the study of identity, gender, and emotion perception from biological motion

摘要

We present the methods that were used in capturing a library of human movements for use in computer-animated displays of human movement The library is an attempt to systematically tap into and represent the wide range of personal properties, such as identity, gender, and emotion, that are available in a person's movements. The movements from a total of 30 nonprofessional actors (15 of them female) were captured while they performed walking, knocking, lifting, and throwing actions, as well as their combination in angry, happy, neutral, and sad affective styles. From the raw motion capture data, a library of 4,080 movements was obtained, using techniques based on Character Studio (plug-ins for 3D Studio MAX, AutoDesk, Inc.), MATLAB (The MathWorks, Inc.), or a combination of these two. For the knocking, lifting, and throwing actions, 10 repetitions of the simple action unit were obtained for each affect, and for the other actions, two longer movement recordings were obtained for each affect We discuss the potential use of the library for computational and behavioral analyses of movement variability, of human character animation, and of how gender, emotion, and identity are encoded and decoded from human movement.