Automated detection of viewer engagement by head motion analysis

Güler, Uğur
Measuring viewer engagement plays a crucial role in education and entertainment. In this study we analyze head motions of the viewers from video streams to automatically determine their engagement level. Due to unavailability of a dataset for such an application, we have built our own dataset. By using face detection system, the head position of viewer is obtained throughout the video for each frame. Then, using these positions, we analyze and extract some features. In order to classify the data, we employ both Random Forest and Support Vector Machine (SVM) with extracted parameters. User engagement detection is performed using the employed model and the results indicate accuracy of 89.4% and recall of 90.9% on our dataset with Random Forest.