基于Adaboost算法的硬件实时人脸检测

实现一个硬件人脸检测系统，该系统工作频率为70 MHz，能够检测一幅256×256的图像中任意位置、任意大小和任意数目的人脸，检测速度为每秒35帧。系统的检测精度为85%，误检率为1.5×10-6。为实现高速人脸检测，在硬件系统架构上做出如下3点创新：实现积分图像和积分平方图像的硬件实时计算，弱分类器特征值计算的深流水线实现，采用并行多内存组织结构。

Real-time Hardware Face Detection Based on Adaboost Algorithm

This paper presents a hardware implementation of face detection system that can detect any number of faces with any size at any position in a 256×256 image, at a speed of 35 fps when the FPGA works at 70 MHz. The detection system can achieve a detection rate of 85% with a false alarm rate of 1.5×10-6. To achieve high speed face detection, three contributions are made in the hardware architecture: real-time hardware implementation of the calculations of integral image and integral squared image, deep pipeline architecture of weak classifier implementation, and parallel multi-memory organization.