Real-Time Multi-Task Facial Analytics with Event Cameras

Event cameras, unlike traditional frame-based excel in detecting and reporting changes light intensity on a per-pixel basis. This unique technology offers numerous advantages, including high temporal resolution, low latency, wide dynamic range, reduced power consumption. These characteristics make event cameras particularly well-suited for sensing applications such as monitoring drivers or human behavior. paper presents feasibility study the using multitask neural network with real-time facial analytics. Our proposed simultaneously estimates head pose, eye gaze, occlusions. Notably, is trained synthetic camera data, its performance demonstrated validated real data driving scenarios. To compensate global motion, we introduce novel integration method capable of handling both short long-term dependencies. The our analytics quantitatively evaluated controlled lab environments unconstrained results demonstrate that useful accuracy computational speed achieved by to determining pose relative eye-gaze direction. shows neuromorphic can be realized are edge/embedded computing deployments. While improvement ratio comparison existing literature may not favorable due event-based vision approach employed, it crucial note research focuses specifically vision, which distinct advantages over RGB vision. Overall, this contributes emerging field systems highlights potential networks combined subjects. techniques applied practical driver systems, interactive human-computer behavior analysis.