Note: All the informations presented below are published on Ferreira et. all (2012).Activity and Emphasis, from the first 60 seconds of video recordings of user interaction to predict the user's experienced task difficulty.The main proposal is to extract features using simple and computationally inexpensive video processing techniques. As a first approach, we extracted participant’s motion using the difference between frames. Video of the user's interaction and the frame difference processing result Activity - measures the participant’s activity level. This variable is the fraction between the number of motion frames and the number of total frames of interaction time.Emphasis - measures the variation of motion's energy and frequency. The variable is a sum of the Fourier transform applied to the frame difference signal and the signal’s standard deviation. For this initial approach only 60 seconds of video recording were used.Then, we developed a classification model using the two computed variables: Activity and Emphasis. Research Hypotheses Hypothesis 1: Activity is correlated with experienced difficulty.Hypothesis 2: Emphasis is correlated with experienced difficulty. ResultsThese two variables, that measure different features of the signal, were found to be correlated with the experienced difficulty (table 1). As such, the two hypotheses were confirmed: Activity is negatively correlated while Emphasis is positively correlated with the experienced difficulty.Activity is less important than Emphasis to justify the experienced difficulty. These two variables justify 46.6% of the variance of task difficulty. In sum, motion tends to be lower (Activity) and more irregular (Emphasis) with the increase in task difficulty. ReferencesPentland, A. 2006. "A Computational Model of Social Signaling". Proc. ICPR'06, IEEE. Kapoor A., and Picard, R. W., 2005. "Multimodal affect recognition in learning environments". Proc 13th annual ACM international conference on Multimedia, ACM, 677-682. |