摘要
连续采集多帧视频图像,利用帧间滤波法建立背景图像,并通过改进背景差分算法尽可能地提取完整的目标轮廓,并用链表法表示轮廓特征。提取目标轮廓的HOG特征,通过SVM分类器进行分类,研究分析不同的人体异常行为。该方法可以有效识别出快速移动、弯腰行走、跌倒、跳跃、长时间停留等异常行为。
The multi-frame video images are continuously acquired,and the inter-frame filtering method is used to establish the background image.Through the improvement of the background subtraction algorithm,the complete target contour is extracted as far as possible,and the linked list is used to represent the contour features.The HOG features of the object contour are extracted and classified by the SVM classifier for the study and analysis of different human abnormal behaviors.This method can effectively identify the abnormalities such as fast moving,bending walking,falling,jumping,long stay,etc.
作者
李明
杨凯
王军
杜文凯
Li Ming;Yang Kai;Wang Jun;Du Wenkai(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221000,China)
出处
《实验技术与管理》
CAS
北大核心
2018年第11期38-41,共4页
Experimental Technology and Management
基金
江苏省教育科学"十二五"规划重点课题(B-b/2015/01/032)
中国矿业大学"十三五"品牌专业建设项目(电子信息工程)(24180017)
中国矿业大学教育教学改革与建设课题(2017YB12)
关键词
异常检测
行为分析
HOG特征
图像处理
abnormal detection
behavior analysis
HOG feature
image processing
