一种适应运动前景的快速行人检测算法  

Rapid Pedestrian Detection Algorithm Based on Foreground

作　　者：鲁超[1] 滕国伟[1] 邹雪妹[1] 安平[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《电视技术》2015年第1期113-116,132,共5页Video Engineering

基　　金：国家自然科学基金项目(61172096;61271212);新型显示教育部重点实验室开放课题资助项目(P201104);上海市科委重点资助项目(12511502502)

摘　　要：针对智能监控视频中行人的运动特性和基本HOG加SVM人体检测算法的特点,将运动目标检测算法融入基本的HOG加SVM人体检测算法。首先,利用运动目标检测算法确定运动区域矩形窗,并扩大调整矩形窗尺寸获得ROI窗口;然后,根据运动区域尺寸与训练样本尺寸的差距调整ROI窗口的首级窗口缩放因子,并对ROI窗口进行人体检测。实验结果表明,本文算法优于基本的HOG加SVM人体检测算法,具有良好的实时性和适应性。Taking advantage of features of pedestrian in intelligent surveillance video and the characteristics of the fundamental HOG plus SVM human detection algorithm, the moving target detection algorithm is added into the fundamental HOG ＋ SVM human detection algorithm. Firstly, moving target detection algorithm is used to determine the rectangular foreground area windows. Then, the prospects of background are added into the rectangular foreground area windows to gain ROI windows by expanding the size of rectangular windows. Secondly, according to the gap between the size of moving region and that of training samples, the first level of the scaling factor of ROI windows is resized and ROI windows are detected. The experimental results show that this algorithm is superior to the fundamental HOG plus SVM human detection algorithm, with good real-time performance and adaptability.

关 键 词：运动目标检测 ROI窗口 窗口缩放因子 行人检测 

分 类 号：TN911.73[电子电信—通信与信息系统]