基于形状上下文的公共场所行人检测系统  

Pedestrian Detection for Public Place based on Shape Context

作　　者：丁辰瑜[1] 葛万成[1] 陈康力[1] 

机构地区：[1]同济大学电子与信息工程学院,上海200092

出　　处：《通信技术》2012年第4期46-49,共4页Communications Technology

基　　金：德国罗德与施瓦茨公司资助

摘　　要：交通规划中对于慢行交通对象流量的检测精度要求越来越高。虽然引入了一些基于智能监控的手段,但是在取得检测结果的速度和精度上仍然有较大的提高余地。为此论文提出了一种基于视频监控和图像识别的公共行人数量检测手段,利用形状上下文和广义Hough变换技术来提高检测结果的准确度和速度。测试结果表明,该检测方法优于传统的基于SIFT特征的检测方法,在大多数中低人流密度的场景下有较高的精度。另外该方法使用的样本数量较少,训练时间也较短,适用于需要进行实时处理的场合。There is an increasing requirement of the critical dataset such as the number of pedestrian in public place for the transportation planning.Although some methods based on visual detection are introduced,the detection results are not accurate enough.This paper presents a system based on visual detection and pattern recognition.The experiment result shows that this system based on Shape Contexts and Generalized Hough Transform is better in performance that the traditional system based on SIFT features,and is fairly suitable for unstuffed places and real-time conditions.

关 键 词：行人检测 广义HOUGH变换 形状上下文 

分 类 号：TN929.5[电子电信—通信与信息系统]