Abstract:

Aiming at the problem of poor real-time performance of most embedded face detection systems, the optimized face detection algorithm and software/hardware cooperative processing method are used to accelerate the face detection. Based on the ZYNQ SoC architecture, the YCbCr skin color space algorithm is used to accelerate the skin color region in the FPGA part, and then the optimized Adaboost algorithm and the Phash algorithm are used to complete the face detection and tracking in the dual-core ARM. It outputs the detected face. Experiments show that the proposed optimized face detection algorithm is more efficient than the traditional Adaboost face detection algorithm, and the reasonable hardware and software co-processing can not only speed up the face detection rate, but also reduce the system hardware resource consumption and cost.

Key words: face detection, ZYNQ, skin color segmentation, synergetic processing of hardware and software, perceptual Hash algorithm

摘要：

针对目前大多数嵌入式人脸检测系统实时性差的问题，通过优化的人脸检测算法和软硬件协同处理方式达到加速人脸检测的目的。基于ZYNQ SoC架构下，利用YCbCr肤色空间算法在FPGA部分加速提取肤色区域，利用优化的Adaboost算法与Phash算法在双核ARM中完成人脸检测与追踪，输出检测到的人脸。实验表明，提出的优化人脸检测算法相比传统的Adaboost人脸检测算法更具实时性，并且通过合理的软硬件协同处理也可以加快人脸检测速率，同时减少系统硬件资源消耗量从而降低成本。

关键词: 人脸检测, ZYNQ, 肤色分割, 软硬件协同设计, 感知哈希