摘要
Adaboost算法采用由弱到强的级联型分类器用以快速检测人脸。但在实际应用中计算量巨大。在PC机上用纯软件实现该算法得到的目标检测速度也难以达到实时。本文论述了一种采用像素积分计算阵列的人脸检测系统,能够对图像像素进行流水运算处理以达到提升检测速度的效果,并在Virtex5系列FPGA上实现。通过该并行系统对单幅352x288的图像进行人脸检测,其速率可以达到50帧/秒,可以满足工业应用的实时性要求。
Adaboost algorithm applies the enhanced cascaded classifiers to detect human face fast. It requires huge computation volume. Even the detecting speed is not acceptable for implementing this algorithm with pure software on PC platform. This article presents an architecture composed by pixel integration computing arrays. Through the pipeline pixel process to the image and implement the system on Virtex5 FPGA platform, the speed is boosting obviously. Take the face detect test with resolution of 352x288 to the system, the result can be 50 fps, which could meet the requirement of the industry application field.
出处
《微计算机信息》
2009年第29期149-150,209,共3页
Control & Automation
