Research On Prediction And Recognition Of Upper Limb Behavior Intention Based On Inertial Sensor

People with hand disabilities account for a large proportion of the disabled group.In order to restore the self-care ability of the handicapped as much as possible,a variety of handicapped mechanical artificial hands have been developed to replace real hands to help the disabled perform daily tasks.The upper limbs of the human body are the "traction tools" of the hands,and the daily operation of the hands is closely related to the movement of the upper limbs.Based on low-cost inertial sensors,this article explores the laws of upper-limb movement and its daily behavior.It mainly studies the two major contents of upper-limb behavior intention prediction and upper-limb behavior recognition,and ensures the high accuracy and real-time performance of prediction and recognition.The summarized upper limb intention information and behavior patterns can be used as a knowledge base for subsequent training of the handicapped manipulator,so as to realize the intelligent control of the handicapped hand.First,the overall design of the upper limb behavior intention prediction and recognition system.A process idea of predicting the behavioral intentions of the upper limbs first,and then identifying specific behaviors based on different intentions is proposed.The hardware selection and communication methods of upper limb movement data acquisition equipment are introduced;the data acquisition upper computer software is designed and developed,including data display,storage and other functional modules.Secondly,the calculation of upper limb pose is studied.Combining the respective advantages of the attitude angle calculation method based on the gyroscope and the attitude angle calculation method based on the accelerometer and the magnetometer,the Kalman filter method is used to fuse the first two methods to realize the accurate estimation of the attitude angle.A mathematical model of the upper limb linkage is established,the coordinate system direction of each joint is determined,and the position of the upper limb end is accurately calculated by forward kinematics,which lays the foundation for upper limb behavior research.Then,the behavioral intention prediction of upper limbs was studied.By analyzing the hand angular velocity data of upper limb movement,the law of upper limb behavioral intention is obtained,the concept of "preparatory movement" is proposed,and the intention type is defined around the preparatory movement,and the characteristic value that can reflect the intention of upper limb is extracted.Finally,a hidden Markov model is used to predict the behavioral intention of the upper limbs,and the Baum-Welch algorithm and the Viterbi algorithm are used in the modeling phase and the prediction phase,respectively.Then,the behavior recognition of upper limbs was studied.It also analyzes the angular velocity data of the hand of the upper limb movement,summarizes the law of behavior,and proposes the concept of "sub-action sequence".A related algorithm is designed to disassemble the sub-actions of the upper limb behavior to form a sequence of sub-actions.In the recognition stage,the sub-action sequence replaces the sensor time sequence and is applied to the dynamic time warping algorithm to achieve performance improvements in both recognition accuracy and real-time performance.Finally,design experiments to verify the method described above.The posture angle calculation experiment and the upper limb position calculation experiment are designed,and the error analysis of the results is carried out to verify the correctness and accuracy of the pose calculation method.Then,an upper-limb behavior experiment was designed.Based on the motion data samples of multiple subjects,after analysis and calculation,the accuracy results of upper-limb behavior intention prediction and upper-limb behavior recognition were obtained,and the two upper-limb research methods were verified through time-consuming algorithm analysis.Real-time performance.