基于骨骼关键点的异常行为识别及异构平台部署  

Skeleton Key Points-based Abnormal Behavior Recognition and Heterogeneous Platform Deployment

作　　者：杨志芳[1] 李乾 YANG Zhi-fang;LI Qian(School of Electrical and Information Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学电气信息学院,武汉430205

出　　处：《自动化与仪表》2021年第11期49-52,共4页Automation & Instrumentation

摘　　要：目前对于异常行为检测算法较多,但是存在检测精度低、对环境要求高、部署困难等缺点。针对以上存在的问题,该文提出了一种基于骨骼关键点的异常行为检测方法。首先对视频图像预处理,然后通过Associative Embedding算法进行人体关键点的提取。为准确描述人体运动,提出用运动特征矩阵进行人体运动描述,引入SVM分类器利用运动特征矩阵进行行为识别。在HMDB51数据中选取的12类异常行为达到平均91.2%准确率,最后模型在CPU+FPGA异构平台进行加速,达到32 FPS的处理速度。At present,there are many abnormal behavior detection algorithms,but there are some shortcomings,such as low detection accuracy,high requirements for the environment,difficult deployment and so on.In view of the above problems,this paper proposes an abnormal behavior detection method based on skeleton key points.Firstly,the video image is pre-processed,and then the key points of human body are extracted by the Associative Embedding algorithm.In order to describe human motion accurately,this paper proposes to use motion feature matrix to describe human motion,introduces SVM classifier,and uses motion feature matrix for behavior recognition.A total of 12 types of abnormal behaviors with average accuracy up to 91.2%are selected from HMDB51 data,and the model is accelerated on CPU+FPGA heterogeneous platform,with a processing speed reaching to 32 FPS.

关 键 词：异常行为检测 骨骼关键点 SVM 特征矩阵 异构平台加速 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]