Estimation of facial motions in lectures from degraded video considering privacy

Many researchers are trying to estimate the degree of concentration of students on lectures by analyzing the movement of their faces from the video taken of the attendance status. However, taking pictures with a video camera in places like classrooms where attendees are limited has a problem of privacy protection. So, it is impossible to take videos unless all students accept it. If we can analyze the movement of faces from degraded videos that cannot be identified individually, it will be easy for students to accept that they will be taken with a video camera. Therefore, in this paper, we made several low resolution videos from an original video taken of the attendance status and examined how difficult it is to estimate the movement of their faces for these degraded videos. According to some experiments, face detection became difficult gradually due to the degree of smoothing. However, it was showed that if the area of each face can be correctly detected in smoothed videos, we can sufficiently estimate the movement of faces by examining the number of skin color pixel in the area.