On Demand Retrieval of Crowdsourced Mobile Video

The proliferation of mobile cameras has popularized social sharing of videos captured at events such as sports matches, art performances, and lectures. Due to bandwidth and energy constraints, it is often not efficient or desirable to upload all captured videos to a server for sharing immediately after capturing. We propose a pull-based, on-demand mobile video sharing system that allows user to share video captured at an event, with two novel components: 1) a lightweight video metadata extraction algorithm running on the smartphones that considers both temporal and spatial features, including sensor (compass) readings and point-of-interest of the content and 2) a video selection algorithm, running on the server, that responds to user queries considering both accuracy of the retrieved video and the upload cost of the smartphones. Our evaluation results show up to four times improvement in upload cost and 52% improvement in subjective quality over three baseline algorithms that consider only either cost or accuracy.