A Real Time Embedded Face Detector on FPGA

Intelligent computing systems demand human computer interfaces that detect and recognize the faces of human users at human interaction speeds. A real-time face detector accurately locates human faces in real-time video streams. While software implementations are effective with regards to accuracy and ease of integration, they suffer from low detection throughput on realtime video signals. In this paper we identify benefits and drawbacks of evolving face detection from software to hardware. We illustrate these tradeoffs by detailing the implementation of an artificial neural network face detector on a midsize FPGA. The prototype achieves 94% detection accuracy at real-time detection rates.