基于ARM的智能视频监控人脸检测系统的设计

Design of intelligent surveillance face detection system based on ARM

针对传统视频监控系统存在的不足,提出了一个基于人脸检测的智能视频监控系统。首先,简述了人脸检测的发展历史和传统人脸检测算法在特征表达与分类器中存在的不足;随后,利用新兴的深度学习网络设计了人脸检测算法。该算法首先使用高斯混合模型提取干净的背景图像,其次采用Goog Le Net网络提取图像的特征,利用长短时记忆网络(LSTMN)对序列的处理能力从特征矢量得到人脸位置及大小。系统运行在基于ARM的树莓派开发板上,采用香港中文大学Mall Street监控数据集对系统进行定性与定量测试,测试性能以查全率和查准率度量,其实验结果分别为55.4%和90.3%,表明该系统具有较强的人脸检测性能。

Aiming at the shortcomings of traditional video surveillance system, an intelligent video monitoring system based on face detection was proposed. Firstly, the development history of face detection and the deficiency of traditional face detection algorithm in feature expression and classifier were briefly introduced. Then, the face detection algorithm was designed by using a new deep learning network. Firstly, a Gaussian mixture model was used to extract a clean background image. Secondly, the features of the image were extracted by Goog Le Net network, and the positions and sizes of faces were obtained from the feature vector by using the Long Short-Term Memory Network（ LSTMN）. The system was run on the Raspberry Pi development board, and qualitatively and quantitatively tested by the Mall Street monitoring datasets of the Chinese University of Hong Kong. The test results were measured by the recall and the precision. The experimental results were 55. 4% and 90. 3%, indicating that the system has an excellent face detection performance.