Classification Of Gait Anomalies By Using Space-Time Parameters Obtained With Pose Estimation

Abstract

Identifying anomalies in people suffering from gait disorders is typically per-formed by invasive methods, which implies attaching equipment to the human body. For instance, electromyography, as well as the use of body markers, are tools used to evaluate pathological gaits. This work presents a non-invasive system for analyzing and classifying normal, hemiparetic, and paraparetic gaits. To this end, we combine computer vision algorithms and artificial intelligence to generate space-time parameters related to the lower limbs’ movement. The proposed methodology consists of capturing RGB images of volun-teers that perform several cycles of the normal, hemiparetic, and paraparetic gaits. Pose estimation models process these images, and intelligent classifiers, based on convolution-al neural networks (CNN) and support vector machine (SVM), and process skeleton gait energy image (SGEI) to achieve characterization and classification of gait, respectively. From the three gait patterns, it is obtained of stride length, cadence, stride width, stride time, gait speed, and angles of the body’s lower extremities. Experimental results show high efficiency in the gait pattern classification, with efficiencies up to 98.57%.