Abstract: This paper transplants the AdaBoost algorithm proposed by Viola from PC platform to TMS320DM642 platform. The transplantation solves the compile problem caused by the difference between PC and DSP platform. The algorithm optimization is proceeding in CCS compiler which is provided by Texas Instruments. The paper optimizes the AdaBoost algorithm by the optimization of C code and the project settings, and a lot of experiments are done to give out the optimization solution. It plays a guiding role in the implementation of real-time feature in embedded DM642 platform.

Key words: DM642, AdaBoost algorithm, face detection, algorithm optimization

摘要： 将Viola等人提出的AdaBoost人脸检测算法由PC机移植到TMS320DM642嵌入式平台，采用EMCV（Embedded Computer Vision Library）移植方案，将PC平台的OpenCV（Open Source Computer Vision Library）视觉库中有关AdaBoost算法的模块修改使之适应嵌入式DSP平台。该移植工作主要解决了OpenCV在嵌入式平台的不兼容问题以及不同体系架构的编译器环境和库文件导致的编译和链接错误。优化工作在TI的编译器CCS中进行，在编译器平台下进行C语言的项目及程序优化，并进行了大量实验数据的对比，给出了优化的方案，对实现系统的实时性有指导性作用。

关键词: DM642, AdaBoost算法, 人脸检测, 算法优化