女子竞技体操跳马运动学变量与裁判打分多元回归模型研究

Multivariate Regression Modelling of Women Artistic Gymnastics Handspring Vaulting Kinematic Performance and Judges Scores

目的:基于体操跳马运动学数据、难度分和最终得分等参数建立跳马预判评分模型并评估预判模型评分的可靠性和准确性,提出备战东京奥运的启示。方法:研究数据来源于2017年全国体操锦标赛女子跳马资格赛中48名运动员的跳马动作。1台JVC PX100高速摄像机用于采集跳马视频数据。运动学数据由有经验编码人员(r=0.96)采用Dartfish二维视频分析解析获得,难度分和最终得分来自于官方比赛成绩公告。样本分为两部分,一部分(n=43人)用于建立跳马数学预判评分模型,另一部分数据(n=5人)用于评估预判模型的可靠性和准确性。使用偏最小二乘法进行回归建模和交叉验证。结果:选择2成分回归模型的拟合优度,预判模型解释方差R2 cal=86.24%,模型验证解释方差R2 val=82.83%。在校准和验证模型中,预测的Y(FS)和参考Y(FS)之间的关系显著(r校正=0.929,r验证=0.910),校准均方根误差和预测均方根误差分别为0.3056和0.3495。结果表明校准数据拟合度较高,模型可以用于描述数据集。难度分、第二腾空时间、倒2步速度、助跑速度、踏板接触时间、倒2步时间和撑马时间是预判评分模型的重要变量。使用5个样本进行模型验证,预测结果显示80%的可靠性和准确性。结论:基于二维视频分析运动学变量和已知的动作难度,可以建立无技术评价的跳马预判评分模型。

Objective:The purpose of this study is to establish a vault judges’scoring prediction model and to evaluate the reliability and accuracy of the predictive model score based on gymnastics vaulting kinematics data,difficulty score and final score.Methods:The object of study was the vaulting of 48 women vaulting athletes in the 2017 National Gymnastics Championship,and a JVC PX100 high-speed camera was used for acquiring video data of horse-vaulting.The kinematics data was encoded by the experienced coder(r=0.96)using Dartfish 2D video analysis software,then the judges’scores were the final score of the official results which also included the difficulty score.The sample was divided into two parts,the first part(n=43)is used to establish the mathematical prediction model,and the other part(n=5)was used to evaluate and predict the reliability and accuracy of the predictive model.Partial least squares regression was used for calibration and cross validation of the model.Results:The goodness-of-fit 2-factor model was selected,the correlation coefficient of explained calibration variance R2 cal=86.24%,and the explained validation variance R2 val=82.83%.The relationship between predicted Y(FS)and reference Y(FS)were significant rcal=0.929,rval=0.910 respectively in the calibration and validation model.The root mean squared error of calibration and prediction was 0.3056 and 0.3495 respectively.This means the calibration data was highly fitted,the model was appropriately describe the calibration data set.The difficulty socre,second flight duration,velocity of last 2nd step,run-up speed,contact duration in springboard,last 2nd step duration,and table support duration were the important dependent variables for the judges’score prediction model.Five samples were used to validate the model,and the predicted results showed 80%reliability and accuracy.Conclusion:The two-dimensional video analysis of kinematics variables and known movement difficulty were sufficient to form a predictive scoring model without technical evaluation.