A large-scale mobile facial recognition system using embedded GPUs
Ahmed El-Mahdy, Radwa Elsersy · High Performance Computing Symposium · 2014
With the rapid growth of mobile phones technology, there is a strong potential for porting server-class applications into the mobile platform. This paper considers the problem of face recognition, and proposes a system that exploits embedded GPUs for allowing for large scale, face recognition. The system provides a novel parallel Fisherface recogniser implementation that allows for a configurable and scalable number of parallel threads. The system also exploits the large secondary storage available, to further scale the recognition database. An initial prototype implementation allows up to 1024 face database size with a recognition time of 210ms on iPhone5 system. The GPU resulted in nearly 10x improvement in performance and 7x reduction in power.