基于3D骨骼数据的课堂个人行为识别研究  

Research on Class Individual Action Recognition Based on 3D Skeleton Data

作　　者：徐苏杰 高尚[1] 张梦坤 朱乐俊 XU Sujie;GAO Shang;ZHANG Mengkun;ZHU Lejun(Jiangsu University of Science and Technology,Zhenjiang 212003)

机构地区：[1]江苏科技大学,镇江212003

出　　处：《计算机与数字工程》2022年第8期1662-1666,1685,共6页Computer & Digital Engineering

摘　　要：课堂作为教育学生的平台,有着至关重要的作用。学生在课堂上的表现很大程度上影响着他们对知识的接受度,所以课堂行为识别对教学有着重要的意义,不仅能提高教学的效率,也能提高教学的质量。论文使用Kinect传感器在课堂环境中获取深度图像并提取学生骨骼的3D坐标数据,定义各个关节点与髋关节的欧式距离,各个关节点与髋关节夹角的余弦值,每一帧的关节点与第一帧的关节点的欧式距离,每一帧的关节点与第一帧的关节点夹角的余弦值为特征,接着用PCA算法进行降维,最后采用基于多核函数的SVM非线性分类器进行分类,在公测数据集和实测数据集上都有较好的识别准确率。As a platform to educate students,classroom plays a vital role.The performance of students in class greatly affects their acceptance of knowledge.Therefore,classroom behavior identification is of great significance to teaching,which can not only improve the efficiency of teaching,but also improve the quality of teaching.This article uses the Kinect sensor in the classroom environment for students depth image and extract the skeleton of a 3D coordinate data,defines the key points and Euclidean distance of hip joint,the key points and the cosine of the angle of the hip,key points of each frame with the first frame of the key points of Euclidean distance,the key points of each frame with the first joint point cosine of the angle of characteristics of frame,then uses PCA dimensionality reduction algorithm.Finally,the SVM nonlinear classifier based on multi-kernel function is used to classify the data,and the recognition accuracy is better in both the open and measured data sets.

关 键 词：Kinect传感器 课堂行为识别 SVM 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]