GAIT Analysis: Methods & Data Review
Authors
Abstract
The physiotherapists analyse gait patterns to recognize normal and pathological gait movements. The difficulty of the gait analysis can be high due to patterns’ complexity and variability. The gait patterns are affected by the characteristics of the individual (gender, age, weight and height) and the walking speed. In this paper, it is proposed a Machine Learning (ML) algorithm to generate knee angle patterns in sagittal plane, which is one of the joints used during the walk. The ML algorithm can generate a specific reference of normal knee pattern depending on individual’scharacteristics and walking speed. This specific reference provides a personalized gait analysis. To this end, three ML approaches are compared: an Artificial Neural Network (ANN), an Extreme Learning Machine (ELM) and a Multi-output Support Vector Regression (MSVR). Using the patterns of healthy people collected by a vision system, authors show that ELM outperforms ANN and MSVR. The ELM can generate specific reference knee patterns for female and male gender. The reference knee patterns generated by the ELM can be used by a gait analysis system which the team proposes to future work. The main goal of the gaitanalysis system is to evaluate and classify the severity of gait pathology through the comparison of the real pattern of an individual with the specific reference pattern generated by the ML approach. The proposed gait analysis system can help physiotherapy team in the gait pathology diagnosis and evaluation of future pathologies.