Effectively Linking Persons on Cameras and Mobile Devices on Networks

Nowadays, ubiquitous surveillance cameras and wireless networks facilitate to understand online and offline human activities, and considerate attention has been paid to analyze people's activities by associating visual data with network data. It is critical to link mobile devices on networks to their owners in surveillance videos. However, existing works mainly rely on location-related information, while it is invalid to distinguish persons with similar movement behaviors due to the error-prone visual localization of person and wireless localization of devices. To solve this problem, in this paper we propose a framework called PD-Link that uses just one camera and one AP to infer links from mobile devices to their owners. To achieve its goal, PD-Link identifies a feature, person-device interactions (PDI), as the linking cue. We evaluate the performance of PD-Link in realistic scenarios and the experimental results demonstrate the efficacy and effectiveness of our method.