摘要
针对方向梯度直方图(Histogram of Oriented Gradient,HOG)和支持向量机(Support Vector Machine,SVM)对行人检测速度慢、效率低和实时性差的问题,提出了先利用混合高斯模型进行运动检测初判,从而获得候选区,调整候选区域尺寸,对候选区进行HOG特征提取,再送入SVM分类器检测出行人。只对候选区进行HOG特征提取可大幅度减少HOG的计算量,减少特征提取的时间,从而提高了检测的效率,可应用于有实时性要求的场合。
To address the issues of slow speed,low efficiency and poor real-time performance of HOG and SVM in pedestrian detection,a mixture Gaussian model is proposed to make preliminary judgment of pedestrian detection,so as to obtain candidate areas,adjust the size of candidate areas,extract the HOG feature from the candidate areas,and then send it to the SVM classifier for pedestrian detection.Performing HOG feature extraction only for the candidate areas significantly reduces the calculation amount of HOG and reduces the time of feature extraction,thus improving the detection efficiency.The algorithm can be applied in situations with real-time requirements.
作者
谢敏
XIE Min(Tongda College,Nanjing University of Posts and Telecommunications,Yangzhou 225000,China)
出处
《无线电工程》
2020年第10期835-838,共4页
Radio Engineering
基金
南京邮电大学通达学院科研基金项目(XK201XZ19012)。
