Motion Energy Alignment Analysis in Dialogue

Abstract

We study physical alignment in conversations and motion energy (ME) validation. ME is estimated through counts of pixel changes in video regions of interest. ME of interlocutors is quantified with the MEA system, which uses a frame difference algorithm, and it is designed and widely deployed in situations defined to involve a natural dialogue. In the spirit of replication and system validation, we report our efforts to validate motion energy analysis on collaborative dialogues. Within successive windows on ME in video recordings (c. 0.3 seconds duration; 9 frames), and over entire videos without windowing, we correlate the simultaneous ME of paired interlocutors. Then, we compare these values with those that arise where ME for one party is either randomized or reversed. We see expected differences between actual and randomized interlocutors, but not reversals. We demonstrate relations among ME values, ME correlations, and ME validation.