CooperSense: A Cooperative and Selective Picture Forwarding Framework Based on Tree Fusion

Abstract

Mobile crowd photographing has become a major crowd sensing paradigm, which allows people to use cameras on smart devices for local sensing. In MCP, pictures taken by different people in close proximity or time period can be highly similar and different MCP tasks have diverse constraints or needs to deal with such duplicate data. In order to save the network cost and improve the transmitting efficiency, pictures will be preselected by mobile clients and then uploaded to the server in an opportunistic manner. In this paper, CooperSense, a multitask MCP framework for cooperative and selective picture forwarding, was designed. Based on the sensing context of pictures and task constraints, CooperSense structures sequenced pictures into a hierarchical context tree. When two participants encounter, their mobile clients will just exchange their context trees and at the same time automatically accomplish forwarding high-quality pictures to each other via a tree fusion mechanism. Via virtual or real pruning and grafting, mobile clients learn which picture should be sent to the encounter and which one should be abandoned. Our experimental results indicate that the transmission and storage cost of CooperSense are much lower comparing with the traditional Epidemic Routing (ER) method, while their efficiency is almost the same.