Enabling Public Cameras to Talk to the Public

This paper asks: Is it possible for cameras in public areas, say ceiling cameras in a museum, to send personalized messages to people without knowing any addresses of their phones? We define this kind of problem as Private Human Addressing and develop a real-time end-to-end system called PHADE to solve it. Unlike traditional data transmission protocols that need to first learn the destination's address, our cameras rely on viewing user's motion patterns, and use the uniqueness of these patterns as the address for communication. Once receiving the wireless broadcast from the cameras, the user's phone can locally compare the "motion address" of the packet against its own motion sensor data, and accept the packet upon a "good" match.

In addition to requiring no data from users, our system transforms the motion patterns into low-dimensional codes to prevent leakage of user's walking behaviors. Thus, a hacker who collects all the broadcast messages would still not be able to infer the motion patterns of users. Real-world experiments show that PHADE discriminates 2, 4, 6, 8, 10 people with 98%, 95%, 90%, 90%, 87% correctness and about 3 seconds constant delay. Since abundant and accurate information can be extracted from videos, PHADE would find applications in various contexts. Extended to localization system and audio guide, PHADE achieves a median error of 0.19m and 99.7% matching correctness, respectively. PHADE can also deliver messages based on human gestures. There is no need to deploy any extra infrastructures or to require users to rent any dedicated device.