Surface electromyography (SEMG) measurement technique for the signal was produced through the contraction of muscles in a human body. The surface electrode is connected on the skin of the muscle. This paper presents an off-line design for the estimation of the actual joint angle of a human leg obtained in performing flexion-extension of the leg at slow and high speeds movement. The design is composed of two phases. The first is measurement of real EMG signal of human leg performance by using SEMG technique and processing this signal with filtering, amplification and then normalized with maximum amplitude. The second phase is to design an artificial neural network (ANN) and train it to predict the joint angle from the parameters extracted from the SEMG signal. Three main parameters of EMG signal are used in the prediction process: Number of turns in a specific time period, duration of signal repetition and amplitude of signal. The design of ANN includes the identification of a performing human leg EMG signal with two speed levels (slow-fast) and estimation of knee joint angle by recognition process depending on the parameters of real measured EMG signal. The real EMG signal is measured from full leg-extension to full leg-flexion by (3 sec) with slow motion and (1 sec) at fast motion.
Root mean square (RMS) errors were calculated between the actual angle (measured by the trigonometric formula was applied with any human leg gives real EMG signal measurement) and the angle predicted by the neural network design. This design is simulated by using MATLAB Ver. R2010a, and satisfying results are obtained. That explains the ability of estimation of joint angle for human leg, where the RMS errors are obtained from (0.065) to (0.015) at fast speed leg flexion -extension and from (0.018) to (0.0026) at slow speed leg flexion-extension.