A vision based proposal for classification of normal and abnormal gait using RGB camera

• Published in: Journal of biomedical informatics Vol. 63; pp. 82 - 89
• Main Authors: Nieto-Hidalgo, Mario, Ferrández-Pastor, Francisco Javier, Valdivieso-Sarabia, Rafael J., Mora-Pascual, Jerónimo, García-Chamizo, Juan Manuel
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.10.2016
• Subjects: Accelerometry; Algorithms; Biomechanical Phenomena; Classification; Computer vision; Dynamical systems; Dynamics; Frailty; Gait; Gait analysis; Heel strike; Humans; Movements; Parameters; Pattern recognition; Photography; Proposals; Senility; Signal Processing, Computer-Assisted; Specifications; Toe off; Computer vision; Heel strike; Frailty; Senility; Pattern recognition; Toe off; Gait analysis
• ISSN: 1532-0464; 1532-0480
• DOI: 10.1016/j.jbi.2016.08.003

[Display omitted] •A method for gait parameters extraction and abnormal gait detection from video.•It only requires RGB side view gait video sequences to obtain gait parameters.•Approximation of skeleton joints extraction.•Classification results show that the proposed method has great potential. Human gait is mainly related to the foot and leg movements but, obviously, the entire motor system of the human body is involved. We hypothesise that movement parameters such as dynamic balance, movement harmony of each body element (arms, head, thorax…) could enable us to finely characterise gait singularities to pinpoint potential diseases or abnormalities in advance. Since this paper deals with the preliminary problem pertaining to the classification of normal and abnormal gait, our study will revolve around the lower part of the body. Our proposal presents a functional specification of gait in which only observational kinematic aspects are discussed. The resultant specification will confidently be open enough to be applied to a variety of gait analysis problems encountered in areas connected to rehabilitation, sports, children’s motor skills, and so on. To carry out our functional specification, we develop an extraction system through which we analyse image sequences to identify gait features. Our prototype not only readily lets us determine the dynamic parameters (heel strike, toe off, stride length and time) and some skeleton joints but also satisfactorily supplies us with a proper distinction between normal and abnormal gait. We have performed experiments on a dataset of 30 samples.