在人机协作过程中,由于光照条件等环境因素和机器人设备摆放等遮挡原因,导致使用基于视觉的运动捕捉设备对人体运动进行捕捉时时间序列的轨迹数据有缺失,进而导致意图识别不准确,增加了机器人运动的不确定性.因此,提出了一种基于支持向量回归(support vector regression,SVR)和长短期记忆(long short-term memory,LSTM)的人体上肢运动时间序列轨迹缺失补偿方法.采用网格搜索法对SVR模型中的参数进行优化来完善历史样本数据集,结合长短期记忆网络对短、长时间序列轨迹缺失的预测补全更精确的优势,将SVR模型补全的历史样本数据集输入LSTM模型训练,进一步降低补偿误差.实验结果表明,在三维空间350 mm的运动尺度范围内,轨迹缺失程度为10％时,SVR-LSTM模型补偿轨迹的平均误差是0.14 mm;轨迹缺失程度为30％时,SVR-LSTM模型补偿轨迹的平均误差是0.47 mm.
Human Upper Limb Motion Occlusion Trajectory Compensation Method Based on SVR-LSTM
In the process of human-computer cooperation,the usage of vision-based motion capture e-quipment could cause the lack of human upper limb time series motion trajectory due to environmental factors such as lighting conditions and site restrictions such as the placement of robot equipment,resulting in inaccurate intention recognition,and thus increasing the uncertainty of robot motion.Therefore,an oc-clusion time series trajectory of human upper limb motion compensation method was proposed based on support vector regression (SVR) and long short-term memory (LSTM).The grid search method was used to optimize the parameters in the SVR model to improve the historical sample data set.Combining the ad-vantages of the LSTM in more accurate prediction and completion resulted for short and long-time series track missing,the historical sample data set supplemented by the SVR model was input into the LSTM model training to further reduce the compensation error.The experimental results showed that the average error of the SVR-LSTM model compensation trajectory was 0.14 mm in the range of 350 mm motion scale in three-dimensional space when the trajectory missing degree was 10％.When the degree of track miss-ing was 30％,the average error of SVR-LSTM model compensation track was 0.47 mm within the range of 350 mm motion scale in three-dimensional space.
occlusion trajectorytime seriesintent identificationtrajectory compensationSVR-LSTM model
万方数据
维普
