The MPI Emotional Body Expressions Database for Narrative Scenarios

Abstract

Emotion expression in human-human interaction takes place via various types of information, including body motion. Research on the perceptual-cognitive mechanisms underlying the processing of natural emotional body language can benefit greatly from datasets of natural emotional body expressions that facilitate stimulus manipulation and analysis. The existing databases have so far focused on few emotion categories which display predominantly prototypical, exaggerated emotion expressions. Moreover, many of these databases consist of video recordings which limit the ability to manipulate and analyse the physical properties of these stimuli. We present a new database consisting of a large set (over 1400) of natural emotional body expressions typical of monologues. To achieve close-to-natural emotional body expressions, amateur actors were narrating coherent stories while their body movements were recorded with motion capture technology. The resulting 3-dimensional motion data recorded at a high frame rate (120 frames per second) provides fine-grained information about body movements and allows the manipulation of movement on a body joint basis. For each expression it gives the positions and orientations in space of 23 body joints for every frame. We report the results of physical motion properties analysis and of an emotion categorisation study. The reactions of observers from the emotion categorisation study are included in the database. Moreover, we recorded the intended emotion expression for each motion sequence from the actor to allow for investigations regarding the link between intended and perceived emotions. The motion sequences along with the accompanying information are made available in a searchable MPI Emotional Body Expression Database. We hope that this database will enable researchers to study expression and perception of naturally occurring emotional body expressions in greater depth.

Toward an efficient body expression recognition based on the synthesis of a neutral movement

“…Notices that the SIGGRAPH database includes the largest range of movements: jump, run, kick, walk, punching and transitions between these motions. [29] 1443 11 Figure 7 shows the influence of the size of the training set on the performance of the three classifiers used in our method. We compared the performance of our method with Support Vector Machine (SVM) with χ 2 kernel, Random Forest with 100 trees and 2-Nearest neighbor based on Euclidean distance.…”

“…Many existing approaches focus on very specific actions, as often done in psychology studies. Many papers have focused on the locomotion [5,14,23,24,24], some on knocking actions [6,11], on artistic performance [25], or on talking persons [29]. Body movements are characterized by a high dimensional configuration space with many interrelated degrees of freedom.…”

Generic Body Expression Recognition Based on Synthesis of Realistic Neutral Motion

“…Lastly, model-based approaches that propose efficient analysis on the skeleton motion are still preferred [23], in comparison to brute force approaches letting a network explores a very large movement dataset. Thus, many articles focus on the most common motions such as walking [25], [25]- [30], action of knocking [31], [32], talking persons [33], or other artistic performance [34]. These methods are adapted on specific motions, but they fail when the challenge is to recognize body expressions from different scenarios.…”

RAMAS: Russian Multimodal Corpus of Dyadic Interaction for studying emotion recognition

“…Ten semi-professional actors (5 men and 5 women, 18-28 years old, native Russian speakers) participated in the data collection. Semi-professional actors are more preferable than professional actors for analyzing movements in emotional states, as professional theatre actors may use stereotypical motion patterns [13,17].…”

“…The interactive emotional dyadic motion capture database (IEMOCAP) [12] contains audio-visual and motion data for faces and hands only, but not for the whole body. The MPI Emotional Body Expressions Database [13] was also collected by means of several channels, yet only motion capture data are available for the research community. Finally, there are well-designed multimodal databases (e.g.…”