Multi-stream optimization
Methods and systems are provided for transmitting multiple video streams to clients based on individual preferences of the clients and network conditions. A cost-function minimization problem is formulated based on various network parameters. The network parameters (e.g., round-trip time (RTT), available bandwidth, packet loss rate, etc.) may be inputs to this optimization problem, which attempts to find optimal resolutions, bitrates, etc., given the network parameters and the video content. Additionally, by incorporating quality prediction, the cost-function may be based on actual content of video sequences.
1. A method for transmission of one or more video streams to a plurality of clients, the method comprising:
generating one or more content metrics for a sequence of video frames based on content contained in the video frames;
estimating a quality function for the sequence of video frames using a regression model and the one or more content metrics;
determining parameters for encoding the sequence of video frames by minimizing a joint delay and distortion cost function, wherein distortion cost is inversely related to the estimated quality function, and delay cost is based on round trip transmission time; and
encoding the sequence of video frames using the determined parameters,
wherein estimating the quality function includes computing a conditional probability density function based on the content metrics, the content metrics including at least bitrate and resolution,
wherein determining the parameters for encoding the sequence of video frames includes determining the bitrate and resolution that maximizes the conditional probability density function.
2. The method of claim 1 , further comprising: determining an amount of bandwidth to allocate to forward error correction for each of the one or more video streams.
3. The method of claim 1 , wherein the regression model is a Gaussian mixture model.
4. The method of claim 1 , wherein the quality function is a structural similarity metric that is approximated as a combination of average motion vector magnitude, average spatial prediction error, and average block variance for the sequence of video frames.
5. The method of claim 4 , wherein the average spatial prediction error is computed as the difference between a pixel value and the average value of one or more neighboring pixels in the sequence of video frames.
6. The method of claim 1 , wherein the quality function for the sequence of video frames is estimated using the regression model, the content metrics, and trained model parameters corresponding to the regression model.
7. The method of claim 1 , wherein the parameters for encoding include one or more of resolution, bitrate, and forward error correction rate.