首页 | 本学科首页   官方微博 | 高级检索  
     

行人闯红灯事件视频检测算法与实现
引用本文:张高煜. 行人闯红灯事件视频检测算法与实现[J]. 黑龙江电子技术, 2013, 0(12): 53-56
作者姓名:张高煜
作者单位:上海金融学院信息管理学院,上海201209
基金项目:上海教委科研创新基金(11YZ233);上海金融学院重点课程建设基金支持(2012LINUX)
摘    要:在回顾了现有的基于视频的行人检测研究成果之后,提出了一种基于历史信息和目标快速特征提取的行人闯红灯事件检测算法,并实际应用于位于上海内首个行人闯红灯报警系统中:首先对实时采集到的每时间单位的视频图像序列进行逐帧差分以过滤背景信息;接着对差分后的图像序列添加历史信息;进而对此图像中兴趣区域,采用线扫描的方法来对运动目标方向、轮廓、轨迹等特征进行快速提取、分析与识别,实现实时的人车分离辨识、行人位置确认和运动趋势预测.

关 键 词:历史信息  单目视觉  行人快速检测

Pedestrian red light running event video detection algorithm
ZHANG Gao-yu. Pedestrian red light running event video detection algorithm[J]. , 2013, 0(12): 53-56
Authors:ZHANG Gao-yu
Affiliation:ZHANG Gao-yu (School of Information Management, Shanghai Finance University, Shanghai 201209, China)
Abstract:In video based pedestrian detection, common approaches applied principle component analysis, Haar wavlet, or Histogram of oriented gradient for feature collection, then neural network, support vector machine, adaBoost or K-mean were used for feature classification. However, the main constraints of them is the computing cost and time issue in real-time pedestrian detection. This paper proposed a novel algorithm based on the history information in video clips and human feature fast retrieving. Based on the intensity difference in sequential frames and with self-adaptive threshold, the region of interested are scanned in horizon and vertical direction for human edge, moving direction and track recognition. Human and vehicles are separated for human moving direction prediction. The approach is used in Eshan Road, Dongfang Road, Shanghai for pedestrian alerting in red light time. The error rate detection is below 17%.
Keywords:video history information  mono-vision detection  fast pedestrian detection
本文献已被 维普 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号