基于混合高斯和HOG+SVM的行人检测模型  

Novel model of pedestrian detection based on Gaussian mixture model and HOG+SVM

作　　者：龚露鸣 徐美华[1,2] 刘冬军 张发宇[1] GONG Luming;XU Meihua;LIU Dongjun;ZHANG Fayu(Microelectronic R&D Center,Shanghai University,Shanghai 200072,China;School of Mechatronic Engineering and Automation,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学微电子研究与开发中心,上海200072 [2]上海大学机电工程与自动化学院,上海200444

出　　处：《上海大学学报（自然科学版）》2018年第3期341-351,共11页Journal of Shanghai University:Natural Science Edition

基　　金：国家自然科学基金资助项目(61376028)

摘　　要：为了提高行人检测系统的检测率,提出了一种基于混合高斯背景建模结合方向梯度直方图(histogram of oriented gradient,HOG)+支持向量机(support vector machine,SVM)的行人检测模型.首先,采用混合高斯模型进行前景分割,有效提取出运动目标区域;然后,在行人识别部分通过缩小检测窗口尺寸来降低HOG特征维数;另外,利用误识别区域,对样本库的信息进行二次更新,以优化SVM分类器;最后,以随机视频帧为测试样本进行模型性能验证.结果表明,在保证检测率和检测速率的情况下,该混合高斯结合HOG+SVM模型的误检率仅为4%,说明该模型能够在复杂场景下实时准确地进行行人检测.To improve detection rate of pedestrian detection system,a novel pedestrian detection model by combining the Gaussian mixture background model and histograms of oriented gradients(HOG)plus support vector machine(SVM)is proposed.First,foreground segmentation is done using the Gaussian mixture model(GMM)to extract moving target areas.In the recognition of pedestrians,the dimension of the HOG descriptor is reduced by resizing the size of the detecting window.In addition,an error recognition region is used to re-update the information of sample dataset to optimize the SVM classifier.Performance of this model is evaluated with the test frames randomly chosen from a video taken in a realistic scene.The results indicate that GMM and HOG+SVM can ensure accuracy and speed of detection,and limit false detection rate to 4%.Real-time and accurate pedestrian detection in complex scenes are achieved.

关 键 词：行人检测 混合高斯模型 区域提取 梯度方向直方图 降维 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]