A new event-driven Dynamic Vision Sensor based Physical Unclonable Function for camera authentication in reactive monitoring system

Abstract

Surveillance footage has become an integral part of law enforcement as video cameras become ubiquitous, affordable and more reliable. Dynamic vision sensor (DVS) emerges as a new sensing technology that outsmarts existing static CMOS image sensors in vision-enabled traffic monitoring, assisted living and high-speed target tracking for its low latency, high temporal resolution and wide dynamic range under uncontrolled illumination. Instead of recording a steady stream of snapshots taken at a fixed rate, DVS responds only to temporal contrast and records only sparse asynchronous address-events with precise timing information. However, the accountability of the footage captured is incomplete if the cue is triggered by an unidentified device. One effective way to eliminate the anonymity is to build a random oracle out of the DVS sensor and use its authenticity as a root of trust to protect the integrity of the footage. In this paper, we present the first ever event-based physical unclonable function (PUF) for DVS camera identification and secret key generation in reactive monitoring system. A non-intrusive PUF response readout scheme is proposed by exploiting the two unique reset switches, one continuous-timed and one self-timed, of DVS pixel to enable simultaneous generation of PUF response with non-disruptive output of asynchronous address-events. Only three transistors are added to each pixel to isolate the PUF response readout and to prevent the spontaneously detected DVS events from interfering with the PUF operation, which is also triggered by the reflectance change in the scene. Our simulation results based on 1.8V 180nm CMOS technology show that the raw response generated by the proposed event-driven DVS-based PUF has near ideal uniqueness of 49.96%, and worst-case reliability of 96.3% and 99.2% for variations of temperature from −35∼115◦C and supply voltage from 1.6∼2.0V, respectively. Its randomness has also been attested by the NIST tests.