AID-R is a low-power, NVIDIA Nano-powered edge compute device which uses human activity recognition to detect assault-related activity. The device can be forward- and rear-mounted in vehicles, thereby forming an ad-hoc video analytic neighborhood watch network by vehicles parked in streets, driveways and parking garages. By doing so at the Smart City edge, the automated video analytics of AID-R respect citizen privacy by not storing video imagery, but can nonetheless immediately alert law enforcement when a critical situation is detected.
AID-R is trained on the human sports dataset supplemented with data that include normal activities such as standing and walking. Context transfer allows the network to infer assault-related activity by generalizing learned human sports activity classes. For example boxing/wrestling in the context of a public urban environment is likely to correspond to an assault activity in progress.
_GFcHZ9V24D.gif?auto=format%2Ccompress&gifq=35&w=400&h=300&fit=min)
Comments