Real-Time Face Detection on a Configurable Hardware System

摘要

We present the design of a real-time face detection system and its implementation on a configurable hardware platform. The system works at 30 frames/second, which is the full frame rate of the video camera. A new detection algorithm was developed that is both robust and extremely efficient in hardware. The image is filtered and then passed through a neural-network-like face detector. Only simple operations are required to calculate the result. The detection accuracy achieved is comparable to other methods from the recent literature. The hardware face detection system is approximately 1000 times faster than the same algorithm running in software on a modern processor.