基于级联网络的智能视频监控下的行人检测  

Pedestrian Detection in Surveillance Video Based on Cascaded Network

作　　者：张姣 肖江剑 周传宏[1] Zhang Jiao;Xiao Jiangjian;Zhou Chuanhong

机构地区：[1]上海大学机电工程与自动化学院,上海200000

出　　处：《计量与测试技术》2018年第8期104-108,共5页Metrology & Measurement Technique

基　　金：浙江省自然科学基金资助项目(No.LR13F020004);国家科技支撑计划(No.2015BAF14B01)

摘　　要：视频中的目标检测与跟踪是实现智能监控系统的关键,在行为分析、辅助驾驶系统、机器人视觉等领域都有着广泛的应用。而对行人的检测与跟踪是其中的关键环节之一。为了在监控场景下准确的进行行人检测,本文深入研究了行人检测算法,并对相关的算法进行优化,设计出了一种基于卷积神经网络(Convolutional neural network,CNN)级联网络的监控视频行人检测系统。在该方案中,引入级联CNN网络在拥挤场景中准确地检测行人,为跟踪行人奠定了基础。本文在论述了以上的创新方法后,釆用神经网络对和人体数据库进行训练后,以实际监控视频作为测试数据进行实验。实验结果表明了本文创新方法具有较好的可行性和有效性。Target detection and tracking in video is the key to the realization of intelligent monitoring system. It is widely used in the fields of behavior analysis,auxiliary driving system,robot vision and so on. Pedestrian detection and tracking is one of the key links. In order to accurately carry out pedestrian detection in the monitoring scene,this paper studies the pedestrian detection algorithm,and optimizes the related algorithms,and designs a monitoring video pedestrian detection system based on the Convolutional neural network（CNN） cascaded network. In this scheme,the cascaded CNN network is introduced to detect pedestrians accurately in crowded scenes,which lays the foundation for tracking pedestrians. After discussing the above innovation methods,this paper uses neural network to train the human database and experiments with the actual monitoring of the British International Airport as the test data. The experimental results show that the innovative method is feasible and effective. And it can meet the high accuracy and real-time performance at the same time.

关 键 词：行人检测 智能视频监控 级联网络 卷积神经网络 

分 类 号：TB9[一般工业技术—计量学]