BS-4-9 HIDDEN MARKOV MODEL FOR GAZE-TRACKING IN NETWORKED VIDEO STEAMING(BS-4. Network Design, Management and Control for Future Networked Systems)

概要

論文の詳細を見る
Network streaming of high-quality video continues to be challenging due to limited bandwidth. One promising approach to reduce transmission rate without degrading visual quality is to forego coding of details away from viewer's central point of focus. Thus, an intelligent bit allocation scheme can allocate more bits to encode the viewer's current spatial region of interest (ROI), while devoting fewer bits in other spatial regions. The difficulty for this real-time bit allocation approach lies in overcoming the network delay between the time the viewer's ROI is detected, to the time effected video is encoded, delivered and displayed. In this paper, using a carefully constructed Hidden Markov Model (HMM), we propose an eye-gaze prediction scheme where the future gaze state of a viewer is predicted. Using the predicted gaze state and prediction confidence level, optimal bit allocation is performed for future frames to circumvent the inherent network delay problem. Our experiments show that the delay could be overcoming, while the prediction accuracy could be improved about 75%, when RTT is 200ms.
2011-02-28