Offloading Image Recognition Processing for Care Robots to FPGA on Multi-access Edge Computing

Due to an aging population, autonomous robots with functions such as detecting falls in elderly individuals, patrolling, and delivering requested items are expected to alleviate the labor burden at nursing care sites. It is challenging to equip robots with computational resources capable of processing these functions due to power constraints, and there is concern that offloading to the cloud will result in significant communication delays. Therefore, we aim to achieve low-latency image processing on a robot using an edge computing method called MEC (Multi-access Edge Computing). This method involves offloading processing from an edge device to a high-performance computing resource via high-speed 5G wireless communication. We also aim to minimize power consumption and latency by leveraging FPGAs, which are expected to serve as promising computational resources for MEC due to their low power consumption and high-speed parallel processing capabilities. In this presentation, we will demonstrate the implementation of the OpenPose-based posture estimation process used for fall detection on an FPGA and its utilization as a computational resource for MEC.