针对人脸识别中实时性的要求，采用FPGA硬件方式实现人脸的实时识别，对传统的LBP算法在硬件实现上存在的问题进行了详细分析，并提出了一种符合硬件数据流处理的LBP优化算法。利用AccelDSP综合工具对该优化算法进行硬件设计，并在Dasal公司的Anaconda卡的IPU FPGA上进行实验验证，满足了人脸识别中实时性的要求。实验结果表明优化后的LBP算法不仅人脸识别率得到了提高，而且在硬件上特征值提取速度是软件上的19倍，能够满足实时性的要求，达到每秒处理100幅人脸图像。%According to the requirements of real-time in face recognition,real-time for face recognition is realized using FPGA hardware, a detailed analysis of traditional LBP algorithm in the hardware implementation issues is made,and put forward a kind of improved LBP algorithm with hardware processing data flow. The hardware design of the optimization algorithm with AccelDSP tools,and experimental verification is made in Dasal company the Anaconda card IPU FPGA,meeting the real-time requirements in face recognition. The experi-mental results show that the optimized LBP algorithm not only increases the face recognition rate,and features in hardware extraction rate is 19 times of the software,which can meet the real-time requirements,processing 100 images per second.
展开▼
