Detecting gait abnormalities using 3D pose estimation

Abstract

The pattern of a person’s limbic movements and behavioral tendencies during locomotion is well-known as a person’s gait, which can get affected due to a medical condition. Persons suffering from neurological disorders often portray aberration in their gait characteristics. These aberrations may involve involuntary movements, pose habits, irregular joint motion, and so on. This project aims to recognize parkinsonian, diplegic, and hemiplegic gaits specifically. These gait abnormalities occur due to Parkinson’s disease, Diplegia, and Hemiplegia, respectively. We capture the 3D human pose patterns across the frames to build a video-level hand-crafted feature set. We designed this feature set while considering different aberrations caused by neurological disorders. That helps us build a machine learning solution that can recognize these abnormal gaits individually and together.