A novel computer vision based gait analysis technique for normal and Parkinson’s gaits classification
Liyun Gong, Jing Li, Miao Yu, Ming Zhu, Ross Clifford
Abstract
Parkinsons disease (PD) can affect a person's gait and potentially lead to some gait impairments (e.g., freezing gait, shuffling gait, etc.) for a PD patient. Analyzing a person's gait characteristics is important for both the early diagnosis and evaluation of their PD. In this work, a novel computer vision based technique is proposed for gait analysis to classify normal or Parkinson's gaits, by using a normal RGB camera. Based on recorded videos of normal gaits, a mask R-CNN, which is a modern deep neural network for objects segmentation, is applied for extracting human silhouettes from video frames. Gait energy images (GEIs) are then obtained from human silhouettes extracted from video clips of normal gaits and processed as features, which are applied to construct a one class support vector machine (OCSVM) model for normal/PD gaits classification. Comprehensive experimental studies show that the proposed technique can successfully classify normal/PD gaits with a high accuracy of more than 97%.