【摘要】 针对人脸检测对高实时性的要求,系统测试了Viola-Jone算法的数据表示(定点/浮点)对检测假阳性和假阴性的影响,检测在逐步降低数据精度的基础上进行。最后在对定点检测评价的基础上提出了一种基于CPU和FPGA混合执行Viola-Jone算法的框架,充分利用FPGA的并行性,对定点表示的Viola-Jone算法进行显著加速处理,极大的提高了人脸检测速度。更多还原

【Abstract】 Aiming at the high real-time requirement of face detection,this paper systematically tests the influence of the data representation(fixed-point/floating-point)of Viola-Jone algorithm on the detection of false positive and false negative,and detects the face on the basis of gradually reducing the accuracy of the data. Based on the evaluation of fixedpoint detection,a frame of Viola-Jone algorithm based on CPU and FPGA is proposed,which makes full use of the parallelism of FPGA to speed up the Viola-Jone algorithm. As a result,the speed of face detection is greatly improved.更多还原