Applying Prediction Techniques to Reduce Uplink Transmission and Energy Requirements in Mobile Free-Viewpoint Video Applications

Abstract

The increased efficiency of video compression algorithms, the improvements registered in reliability, data rates and quality of service of wireless transmission technologies, and the development of mobile multimedia terminals have made possible the implementation of Free-Viewpoint Video (FVV) technology on mobile platforms. The mobile environment however presents several restrictions. Two of these limiting factors being bandwidth constraints and energy availability in battery-operated mobile terminals. This paper looks at the possibility of employing prediction algorithms at the FVV server to anticipate the next viewpoint expected by the user. In doing so, the number of uplink requests is reduced to situations where the estimated view defers from the requested one and the associated transmissions required in retraining the algorithm once this occurs. Simulation results on two different prediction techniques demonstrate that the uplink transmission rate is reduced by up to 96.7% when emulating a conventional FVV usage scenario. Both prediction algorithms infer a substantial decrease in the mobile terminal's power consumption and reduce the network's uplink bandwidth utilization.

Simultaneous Stream Transmission Methods for Free Viewpoint TV: A Comparative Study

“…In linear regression algorithm is implemented at the server for most probable next view prediction. The author in [27] proposed that the requested viewpoint is transmitted to the client. We propose that the next predicted viewpoint is also transmitted along with the actual viewpoint.…”

“…So, this will really help a lot towards fast switching. As this algorithm provides much prediction accuracy [27], so it results in fast view switching, and it is quite suitable for FTV. We have also calculated the regression line based on some data set.…”

“…The calculations in equations (5) and (6) are carried out at the training level in order that the convergence of model functions is done. When these calculations are done and mentioned parameters are set up, the computation complexity at server side is much reduced and there are left only minor additions and multiplications which are performed by the LR algorithm [27], as in equation 3.…”