基于 MEMS 的人体行为特征反演系统设计-《计算机技术与发展》

文章信息/Info

Title:: Design of Human Behavioral Characteristics Inversion System Based on MEMS

作者:: 潘婉苏¹; 2; 李晓风¹; 2; 许金林¹; 李皙茹¹; 程龙乐¹; 2; 1. 中国科学院合肥物质研究院,安徽合肥 230026; 2. 中国科学技术大学,安徽合肥 230031

Author(s):: PAN Wan-su¹; 2; LI Xiao-feng ¹; 2 ; XU Jin-lin ¹ ; LI Xi-ru¹ ; CHENG Long-le¹; 2; 1. Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230026,China; 2. University of Science and Technology of China,Hefei 230031,China

Keywords:: MEMS inertial sensor; posture solution; motion inversion; K-nearest neighbor

摘要:: 人体运动过程较为复杂,传统人体运动姿态识别方法识别动作较为单一,且准确性较低。为了满足人体运动方式检测的需求,设计并实现了基于 MEMS 惯性传感器的人体行为特征反演系统,结合多传感器信息对人体运动姿态进行识别,提高识别的种类及准确率。系统平台采用核心处理器 STM32 搭配传感器模块 MPU9250 组成惯性测量单元,采集人体运动信号,并对采集到的信号进行预处理。研究不同运动模式下的姿态角信息及其变化规律,分析并提取其中最能够反映人体运动的特征参数,构建人体姿态特征库。使用基于 K 最近邻分类算法对样本数据进行训练,建立人体行为特征模型,实现了对不同人体运动姿态的准确识别,同时将姿态识别结果上传至云服务系统;并对人体运动数据进行长时期的跟踪分析,进而反演出用户 24 小时内的行为活动方式。大样本测试结果表明,该系统测量各姿态评估准确性高,重复性好。

Abstract:: The process of human motion is very complex. The traditional human motion posture recognition method has simple recognition action and low accuracy. In order to meet the needs of human motion detection,we design and implement a human behavior feature inversion system based on MEMS inertial sensor. The human motion posture is identified with multi-sensor information to improve the types and accuracy of recognition. The system platform forms an inertial measurement unit with core processor STM32 and sensor module MPU9250 to collect human motion signals and preprocess the collected signals. We study the posture angle information and its change rule under different motion modes,analyze and extract the characteristic parameters which can best reflect human motion,and construct the human posture characteristic database. The classification algorithm based on K nearest neighbor is used to train the sample data and establish the behavioral characteristics model. Different human motion postures are accurately recognized,and at the same time the results of the posture recognition are uploaded to the cloud service system. The human movement data are tracked and analyzed for a long time,and then the behavior patterns of the users within 24 hours are reversed. The results of test on large sample show that the system has high accuracy and repeatability in measuring each posture.