Relevance Metrics for Coverage Extension Using Community Collected Cell-phone Camera Imagery

The rapid increase of mobile phone cameras has enabled users to easily take and share pictures. This has created a potential for mobile device driven sensing of our world at a previously unachieved spatio-temporal granularity, enabling a variety of new applications. The data collection activity is highly uncoordinated and hence, a key issues in effectively using such imagery is understanding the relevance value of each image. Having such a value can not only streamline the resource usage in sharing the image data but also support the development of incentive mechanisms for users to contribute worthwhile data. We discuss the problem of assigning relevance values to images from mobile devices with respect to an application’s existing image data-set. We describe a general information theoretic framework for computing relative relevance and discuss specific value computation for a coverage based metric. We also develop a practical algorithm to compute relevance and describe methods to make our computation scalable to large data sets. Finally, we present our prototype implementation demonstrating our methods  on real world data.