基于单姿态传感组件的游泳动作识别方法  

Swimming Motion Recognition Based on Single Sensing Component

作　　者：肖菊 XIAO Ju(Shanxi Vocational University of Engineering and Technology,Jinzhong Shanxi 030619,China)

机构地区：[1]山西工程科技职业大学,山西晋中030619

出　　处：《电子器件》2022年第5期1264-1271,共8页Chinese Journal of Electron Devices

基　　金：山西省重点研发计划项目(201903D321043)。

摘　　要：针对游泳运动数字化监测系统中游泳动作识别的关键问题,提出了一种基于单姿态传感器与决策树分类思想的游泳动作自主识别方法。首先采用单个无线姿态传感器实时监测游泳运动的三轴加速度与角速度数据,提取游泳动作数据对应的48维统计学特征,并结合实际动作标签生成游泳动作数据库;然后利用训练数据与留一交叉验证方法构建并优化决策树分类器;最后利用分类识别器对测试数据进行游泳动作识别,并根据识别结果对游泳动作切换点进行估计。结果表明:该方法可准确识别四种泳姿的划水和转弯动作,对训练数据和测试数据的平均识别准确率分别可达96.4%和94.8%,平均综合评价指标F-score分别为0.916和0.902;根据分类识别结果也可准确估计游泳动作切换点的时刻,对训练数据和测试数据的平均估计时间误差分别为0.186 s和0.209 s。该方法全面解决了游泳运动中的动作识别问题,为进一步全面评估游泳运动奠定了研究基础。To solving the key problem of swimming motion recognition, a method of autonomous swimming motion recognition is proposed using single sensing component and decision-tree classification. First, the three-axis acceleration and angular velocity of swimming motion were captured, and the corresponding 48-dimensional statistical characteristics were extracted and combined with the actual swimming motion to form a swimming motion database. Then, the training data and leave-one-cross-validation method were used to construct and optimize the decision-tree classifier. Finally, the trained classifier was employed to recognize swimming motion of testing data and estimate the motion-switching point according to its recognition results. Results show that the proposed method can accurately identify the stroke and turning movements of the four swimming styles. The average recognition accuracy of the training data and testing data can reach 96.4% and 94.8%,respectively and the overall evaluation indices are 0.916 and 0.902. Based on the swimming motion identification, the time of the motion-switching point can also be accurately estimated, and the average estimated time error of the training data and testing data is 0.186 s and 0.209 s, respectively. The proposed method comprehensively implements swimming motion recognition and offers a research foundation for further swimming sports evaluation.

关 键 词：姿态传感器组件 决策树分类器 留一交叉验证 游泳动作识别 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP212.6[自动化与计算机技术—控制科学与工程]