Motion Sensing Based On Inertial Sensors

With the development of technology and the improvement of manufacturing tech-nique,mobile devices have ushered in explosive growth.In order to meet the growing needs of people,more and more mobile devices are equipped with a variety of sensors.Inertial sensors are a typical example and they are found in mobile devices such as smart phones,smart watches and smart wristbands.Since inertial sensors can capture a wealth of motion-related information,research on motion sensing based on iner-tial sensors has formed a new wave.In this paper,behavior recognition and motion tracking based on inertial sensors are selected as two entry points.They belong to the category of motion sensing and both need to capture the information of the target subject's movement process through some measurement methods to obtain a meaning-ful impression about the movement.However,their focus is different.For behavior recognition,the meaningful impression about the movement is the category to which the movement belongs.While for motion tracking,it is the quantitative description of certain attributes during the movement that matters.Specifically,the motion sensing mechanism based on inertial sensors is studied in this paper taking human gait motion recognition and rigid body motion tracking as cases.Firstly,based on the research and analysis of three types of gait motion,name-ly,level walking,walking upstairs and downstairs,this paper proposes a novel gait motion recognition method based on inertial sensors.This method extracts the gait motion-related features based on angles and these features are little influenced by user-s' shape and habits,which means this method requires no training from the specif-ic user.Therefore user-independent recognition can be realized.Compared with the high-dimensional frequency domain feature scheme,the proposed scheme can achieve a considerable recognition performance.The average recognition accuracy is 99.5%.Besides,time cost of feature extraction and recognition process can be reduced in con-trast.Hence the proposed scheme is relatively efficient and lightweight.In addition,in the generalization test of behavior recognition,there is no obvious decrease in recog-nition accuracy of the proposed scheme,which means the proposed scheme has good generalization of recognition.Secondly,this paper proposes a rigid body motion tracking method based on in-ertial sensors and a rigid body motion tracking system named MotionTracker is imple-mented based on this method.This method can not only realize single-point trajectory tracking of the rigid body,trajectories of other selected positions on the rigid body can also be restored.By constructing an equivalent model of rigid body motion process,this method equates the complex motion process of a rigid body as a combination of one translation and rotation in three-dimensional space.MotionTracker then analyze the motion process based on this model and the analysis results can be compared with optical tracking systems conveniently.Besides,we propose a Kalman filter based tra-jectory calibration scheme using redundant inertial sensors.By comparing with the measurement results of the optical tracking system,this scheme can partially correct the deviation of the estimated trajectory.Table tennis is chosen as the application sce-nario in this paper.We place the inertial sensor module on the table tennis racket and the table tennis racket is the rigid body to be tracked.MotionTracker can estimate the trajectory of the selected position on the racket and analyze the racket's motion pro-cess with the help of the model mentioned above.Our experiment staff perform three basic table tennis strokes,namely,drive,block and topspin.In the contrast experiment with the optical tracking system named OptiTrack,the average estimated deviation of our translation scalar is about 1cm,the average deviation of the rotation axis is about 18°,and the average deviation of the rotation angle is about 1° in the time interval of 0.04s.In addition,the trajectory restore results of the two systems when we perform topspin are plotted in this paper.So we can observe the motion tracking performance of MotionTracker intuitively.