基于GPU的并行人脸检测方法研究

Research on Parallel Face Detection Method Based on GPU

为了进一步提高人脸检测的精度和速度，提出了一种改进的Viola—Jones人脸检测算法，并在GPU平台上对改进后的算法加以优化。采用新型稀疏特征来进行人脸检测，利用计算机的图形处理单元GPU进行并行加速，并对优化前后的结果进行对比。结果表明，该算法把检测率提高到93．6％，与传统Viola-Jones人脸检测算法相比检测精度有了较大的提升；GPU并行加速实现了对OpenCV1．6～3．2倍左右的加速比，有效提高了人脸检测性能。

In order to improve the accuracy and the speed of face detection,an improved Viola-Jones face detection algorithm is proposed,and the improved algorithm is optimized on the GPU platform.In this paper,a new type of sparse feature is used to detect the human face,and the GPU is used for parallel acceleration.The results before and after optimization are compared.Experiment results show that the detection rate of the improved algorithm increases to 93.6%,and has a greater improvement in detection accuracy compared with the traditional Viola-Jones face detection.GPU parallel acceleration is 1.6to3.2times faster than OpenCV,which effectively improves the performance of face detection.