| In the field of sports somatic science,it is necessary to measure the kinematic information of human limbs in the process of movement.Although modern human motion capture system has been widely used to measure the spatial posture of limbs,its reliability,convenience and accuracy still need to be improved in swimming scene,and it is difficult to automatically segment the swimming phase.Compared with the optical motion capture system,the wearable motion capture system based on inertial sensor has the characteristics of low cost,miniaturization and lightweight,and is not limited by the area.It has become an important measurement equipment in the field of medicine,sports and human-computer interaction.However,due to the complexity of human body movement and the large degree of freedom,the wearable inertial motion capture system is difficult to eliminate the attitude error.At the same time,the existing swimming phase segmentation algorithms are not universal for a variety of phase schemes,so it is necessary to design an effective swimming posture capture and phase segmentation algorithms,so that the swimming motion capture system can measure the swimmer’s whole body posture and motion phase.In view of the above problems,this paper proposes a wearable swimming motion capture system and phase segmentation method.The main research work is as follows:(1)A swimming motion capture system based on micro electro mechanical inertial measurement unit(MEMS-EMU)is constructed,which can quickly and reliably measure the whole body or local body posture of four competitive swimming postures.The magnetometer correction,static detection and zero bias elimination before orientation update algorithm are studied.Body orientation calibration and initial posture estimation of the motion capture system are improved,so that it can be synchronized in the initialization process of the system.In order to improve the convenience of the system.Secondly,the system can use a variety of orientation update algorithms to complete the posture calculation,in order to reconstruct the human biomechanical model;in addition,in order to facilitate the accuracy verification and swimming analysis of the system,the approximate human anatomical position is proposed for the transformation of human posture representation method,in order to meet the needs of pose data visualization.(2)A general swimming phase segmentation method is proposed,which can realize multiple phase segmentation schemes of four competitive swimming postures.Based on the multisensor inertial features of the swimming motion capture system in this paper,a swimming phase segmentation algorithm based on neural network and hidden Markov model is proposed.Secondly,multi-modal features are further extracted to recognize the swimming phase combined with common classifiers.Time tolerance is proposed to evaluate the precision of swimming phase segmentation,and the influence of time tolerance on the segmentation of swimming phase is analysed.On the one hand,the human posture data provided by the swimming motion capture system can be used for the manual annotation of swimming phase,on the other hand,it can be fused with the results of swimming phase segmentation,so that the overall effect of swimming phase segmentation can directly reflect the posture changes of each phase of the four swimming postures.(3)A feature subset selection framework for sensor combination is proposed,which transforms the problem of sensor number selection and combination into the problem of feature subset evaluation and selection,and realizes the automatic optimal configuration of sensors.The improved filtering feature selection method has obvious advantages in the prediction time of phase segmentation algorithm,while the improved embedded feature selection method has better effect in the accuracy of phase segmentation algorithm.The proposed framework can automatically select the specified number of sensors without manual operation,and the influence of the number of sensors on the accuracy of the phase segmentation algorithm is well controlled. |
