Method and apparatus for gait environment classification based on surface electromyography using artificial neural networks

Gait Environment Classification Technology Using Artificial Neural Network Analysis of Surface Electromyography (sEMG) Data

This technology is a method and device that collects muscle activation data for various gait environments using surface electromyography (sEMG) sensors attached to multiple lower limb muscles, such as the rectus femoris, vastus medialis, and tibialis anterior, and uses this data as input for an artificial neural network to estimate and classify the user's gait environment.

Surface electromyography signals are difficult to classify accurately due to their complex patterns and non-linear characteristics, making it challenging to detect transitions in gait environments early enough to control assistive robots effectively.

This technology proposes a method for estimating gait environments by utilizing electromyography profiles from 11 lower limb muscle sites as inputs for an artificial neural network. It can be applied to exoskeleton gait assistive robots and rehabilitation equipment, enabling rapid recognition of changes in the user's gait environment to provide natural and safe assistance.

‍

‍

Key Features:

• Collecting lower limb muscle activation data during gait for each environment using surface electromyography sensors
• Estimating and classifying the gait environment through an artificial neural network using the collected muscle activation data as input
• Configuration involving the attachment of surface electromyography sensors to lower limb muscles, including the rectus femoris, vastus medialis, vastus lateralis, semitendinosus, and biceps femoris
• Configuration for acquiring signals by attaching additional sensors to the tibialis anterior, soleus, medial gastrocnemius, lateral gastrocnemius, and flexor hallucis longus

‍

This invention was developed through the Ministry of Science and ICT's project on early detection algorithms for gait environment transitions based on biosignals using deep learning techniques and support for the unaffected side.