Objective
According to the muscle Hill model to estimate the angle of continuous human movement, existing methods require more parameters, and these parameters are susceptible to the influence of different individuals, so there are often large errors in estimation results.
Methods
Therefore, Back Propagation (BP) neural network model based on features of surface electromyography (sEMG) and the angle of human movement was established in this paper. By studying the role of muscles in joint rotation, appropriate muscle tissues were selected to place EMG sensors, and the model of sEMG features and joint angle was established. For the problem in which sEMG features couldn’t fully reflect all EMG information, an extraction method combining time domain, frequency domain and time-frequency domain features was proposed. Aiming at the problem that the Degrees of Freedom (DOFs) of the forearm lateral movement and the wrist swing were controlled by more muscles, which made the joint angle difficult to predict, a method for correcting the estimation angle error by Kalman filter was proposed. Two DOFs exoskeleton robot was designed, and the established model and prototype were used to perform the tracking experiment.
Results
The average absolute errors of two DOFs are about 17.6° and 6.9°, respectively.
Conclusion
The results suggested that BP neural network model designed couldn’t only achieve uniform velocity tracking of the upper limb, but also ensured that the angular error was within a reasonable value, which met basic requirements for continuous movement estimation of the human body by sEMG.