| 摘要: |
| 目的:采用单个三轴加速度传感器获取身体活动产生的加速度信息,将剧烈程度差别较大的身体活动分为休息(包括静坐、平躺、站立)、步行(包括散步、上楼、下楼)、跑跳(包括快跑和原地跳)3种类型,为评估身体活动水平提供客观依据。方法:使用单个三轴加速度传感器采集44名被试静坐、平躺、站立、散步、上楼、下楼、快跑、原地跳8类身体活动时产生的加速度信息,并进行带通滤波,对滤波后的信号提取信号幅度区域、均值、标准差、三轴相关系数、频域熵等特征量,使用径向基核函数的支持向量机构建身体活动识别模型,并采用交叉验证方法对模型进行验证。结果:采用时频域复合特征子集取得的最高识别率为98.50%,仅采用时域特征子集取得的最高识别率为98.30%,计算分别耗时223 ms和146 ms;窗口大小取5.12 s时识别率最高。结论:单个三轴加速度传感器能够采集丰富的人体活动信息以供识别,建立的径向基核函数支持向量机身体活动类型识别模型能够在自然环境下实时准确划分身体活动类型,能够为身体活动监测提供数据支持。 |
| 关键词: 身体活动识别 支持向量机 加速度传感器 |
| DOI： |
| 投稿时间：2024-03-18 |
| 基金项目: |
|
| Research on the Recognition of Physical Activity Types Based on a Single Triaxial Accelerometer |
| LI Yan,YANG Wenli |
| (School of Sports, China University of Mining and Technology, Xuzhou 221116 , China) |
| Abstract: |
| Objective: A single triaxial accelerometer was used to obtain the acceleration informationgenerated by physical activity. The physical activities with great differences in intensity were dividedinto three types: rest (including sitting, lying flat, standing), walking (including walking, going upstairs,going downstairs), and running and jumping (including fast running and jumping in situ),which provided an objective basis for assessing the level of physical activity. Methods: A single triaxialaccelerometer was used to collect the acceleration data generated by 44 participants during eighttypes of physical activities, including sitting, lying flat, standing, walking, going upstairs, goingdownstairs, fast running, and jumping in situ. The data were then band-pass filtered. The signal amplituderegion, mean value, standard deviation, three-axis correlation coefficient, frequency domainentropy and other characteristic quantities were extracted from the filtered signal. The support vectormachine of radial basis kernel function was used to construct the physical activity recognition model,and the cross-validation method was used to verify the model. Results: The highest recognition rateobtained by using the time-frequency domain composite feature subset was 98.50%, and the highestrecognition rate obtained by using only the time-domain feature subset was 98.30%. The calculationtime was 223 ms and 146 ms, respectively. The recognition rate is the highest when the window sizeis 5.12 s.Conclusions: A single triaxial accelerometer can collect abundant human activity informationfor recognition. The established radial basis kernel function support vector machine physical activitytype recognition model can accurately classify physical activity types in real time in the naturalenvironment, and can provide data support for physical activity monitoring. |
| Key words: physical activity recognition support vector machine accelerometer |
