An Estimator for Rating Video Contents on the Basis of a Viewer´s Behavior in Typical Home Environments

A novel method for predicting the ratings of video content on the basis of a viewer´s behavior in a typical home environment is proposed. Using an input signal provided by a Kinect sensor, it identifies the presence of a viewer by extracting key point trajectories in video sequences of that viewer. It then estimates whether the viewer is gazing at the video content or not on the basis of the viewer´s head pose, which is estimated in two different ways: by a color image-based module and by a depth-image-based module. Results from the two modules are combined to increase robustness. To evaluate the proposed method, a simulation test on TV viewing in a typical living space was conducted. The simulation results suggest that the proposed method can robustly detect the viewer´s gaze and predict his or her rating of the video content.