摘要
Adaboost算法级联目标检测方案使人脸检测向高速实时化迈进了一大步。但是该算法的计算复杂度高,需要存取的数据量非常大,如果采用纯软件的实现方案,会耗费相当多的CPU以及内存资源,难以达到实时检测的要求。本文分析了现有的Adaboost算法硬件架构,对访存效率,耗费的逻辑资源,检测速度等进行了分析,提出了一种基于检测窗口的阵列结构,利用硬件流水特性大大加速了检测过程。本设计通过Xilinx的Spartan3A-DSP型FPGA验证,可满足高清实时人脸检测的要求。
Adaboost algorithm makes it possible to implement real-time face detection system, but the soft algorithm can not work on embedded platform for real-time face detection due to its high computation load and data throughput. This article analyzes the memory efficiency and logic resource of hardware architecture, designs a new cell array architecture using parallel technology. Detection procedure can be greatly speeded up with its multi-pipelines. This design is verified on Xilinx Spartan3A-DSP FPGA. It can work with high resolution stream for real-time face detection.
出处
《电子测量技术》
2009年第1期59-62,共4页
Electronic Measurement Technology
