A Traffic Reduction Method for Crowdsourced Multi-View Video Uploading

The integration of video streams captured by many mobile cameras (contributors) at a crowded event into a multi-view video enables remote viewers to experience the different perspectives of the event. However, because of the resource-constrained nature of wireless networks and the redundant transmission due to the highly correlated streams from multiple mobile cameras at the same event, traffic reduction is necessary to ensure the efficiency of the uploading of crowdsourced video streams. In this paper, we propose a content-based video uploading scheme for crowdsourced multi-view video streaming with the goal of reducing the video traffic from crowdsourced contributors. To achieve this, the proposed scheme uses differential encoding with multiple reference streams by means of packet overhearing. To realize differential encoding across the network of contributors for higher traffic reduction, our scheme combines three techniques: correlation estimation, reference selection, and transmission order determination. First, we utilize the correlation among the contributors based on the content features of the captured video streams using the information-bound reference. Second, in the design of the reference selection that determines the dependencies among the contributors we use two threshold values, determining the number of references for differential encoding at each contributor. Finally, we schedule the transmission order of the contributors to increase the number of differential encoding opportunities within their network. Our evaluation results show that the proposed scheme achieves a traffic reduction of up to 31% with a quality improvement of up to 2.7 dB in the connected network of contributors.