视频监控系统中的行人及其面部侦测研究

Study on pedestrian and face detection in video surveillance systems

采用集成H.264硬件编解码视频处理单元Hi3512来设计视频监控系统,并探讨行人目标的自动侦测问题。在对视频图像进行形态学分析的基础上,利用背景差方法实现运动目标区域的粗提取,通过阴影去除算法实现运动目标的精确定位,再利用连续均值量化变换(Successive Mean Quantization Transform,SMQT)算法实现运动区域灰度图像的增强处理,然后利用SNoW(Sparse Network of Winnows)分类算法实现行人及其人脸部位的侦测。实验结果表明,所采用方法能够自动检测出监控区域的行人目标及其面部信息,可有效地应用于无人值守视频监控场合。

The video surveillance system is designed to investigate the automatic detection of pedestrian goals, which is built by an integrated H.264 hardware codec video processor Hi3512. Based on the morphological analysis of video images, the background subtraction method is used to extract the movement targets. The shadow removal algorithm is used to achieve the precise positioning of the moving targets. To enhance the gray scale image of moving regions, the successive mean quantization transform （SMQT） algorithm is applied, and then, the sparse network of winnows （SNOW） classification algorithm is utilized to detect the pedestrianJs face. Experimental results showed that the pedestrian and facial information can be automatically detected by the proposed method. The method can be effectively employed in unmanned video surveillance application.