- Università di Verona - Dip. Informatica
Tuesday, June 15, 2021
Marker-less human pose estimation (HPE) through RGB camera sensors and deep learning-based SW applications is a trend topic for human motion analysis. Due to latency, privacy, and network bandwidth constraints, there is an increasing interest in applying such a computer vision technique ”at the edge”, by which camera streams are elaborated close to the sensor through low-power embedded boards. Even though some platforms have been proposed to apply HPE on mobile devices, they achieve real-time performance at the cost of low accuracy due to the hardware limitations. In this workshop, we present BeFine, a system that implements the human pose estimation at the edge on an embedded low-power, low-cost CPU+GPU device. BeFine has been customized to detect emergency situations in human working environment (e.g., person fainting in ICE lab) and activate alarm signals.