Research on gait recognition of lower limb exoskeleton robot based on sEMG&IMU feature fusion

Abstract

Aiming at the problems of low accuracy and poor robustness in gait recognition of lower extremity exoskeleton robots in human-computer interaction, a depth residual contraction network recognition method based on the fusion of surface electrosemg (sEMG) and inertial measurement unit (IMU) signals was proposed. Firstly, a new energy kernel feature extraction method was used to extract sEMG signals. Based on the sEMG oscillator model, the sEMG energy kernel phase diagram was converted to gray level map by matrix counting method. Secondly, the IMU signal is denoised and processed graphically. Then, deep residual contraction network (DRSN) was used to recognize sEMG and IMU signals in lower limbs. Finally, experimental hardware was deployed in the wearer's lower limbs, and the algorithm was used to conduct offline and online recognition experiments of three common gaits. Different comparative experiments show that the attention mechanism of DRSN network can significantly improve the classification effect, and the recognition accuracy is improved by 10%-20% compared with single source signal and other feature extraction methods, and finally the recognition accuracy reaches more than 90% through online experiments. The multi-feature fusion network based on energy kernel feature extraction is time-efficient, high-accuracy and robust, and has real-world application value in the field of exoskeleton robotics.