基于Adaboost和PCA的嵌入式人脸识别方法

针对传统的Adaboost算法和主成分分析(PCA)算法用于人脸识别时在环境与姿态等非约束性条件下识别率大大降低以及要求训练样本符合高斯分布的缺陷,提出了一种融合Adaboost和PCA的与或关联决策方法.一方面,在需要安防模式时开启或决策,拒绝近似全部负样本的请求,最大限度保证识别的正确率;另一方面,在需要访客模式时开启与决策,以减少正样本的丢失.在Samsung 2440嵌入式Linux平台上采用该方法进行人脸检测时,基于2种决策方法,分别满足各自阈值.实验结果表明:该方法在嵌入式平台运行稳定,适合推广于智能家居控制与楼宇自动化控制.

Research and implementation of embedded face recognition based on Adaboost and PCA

Aiming at defects of traditional Adaboost algorithm and principal component analysis(PCA) algorithm for face recognition in environment and pose and other nonbinding condition recognition rate is greatly reduced and the requirements of training samples in accordance with the Gauss distribution,a fusion of Adaboost and PCA and or related decision-making method is proposed.On the one hand,when the security mode is needed,the request of the whole negative sample is rejected,and the correct rate of recognition is guaranteed.On the other hand,face recognition in the need to open the visitor mode and decision-making,to reduce the loss of positive samples.On Sumsung 2440 embedded Linux platform,using the method of face detection,based on two decision-making methods,respectively meet the respective threshold.Experimental results show that the method works on embedded platform stably,suitable for promotion in intelligent home control and building automation and control.