Abstract: Objective To develop a non-contact identification method for gait asymmetry in Parkinson's disease based on depth image to assist medical diagnosis and assessment, to avoid the cost, impact on normal life, and complex process of high wear-out sensing equipment. Methods From July to August, 2016, eight patients with Parkinson's disease and ten healthy subjects were collected the gait parameters of walking six meters with Kinect V2.0. The parameters of left and right foot were filtered and clustered. Then similarity matrix algorithm was used to find the difference between healthy subject and patient similarity values. Finally, the recognition effect of this method was verified by Hidden Markov Model. Results The similarity of clustering sequences of left and right foot parameters was less in the patients than in the healthy individuals. There were twelve of 14 data identified in patients, and 35 of 46 in the healthy. Conclusion A non-contact identification method for the asymmetry of gait has been developed based on the parameter clustering results of left and right foot, which is some effective on identifying Parkinson's patients.

Key words: Parkinson's disease, gait symmetry, Kinect, matrix algorithm for string similarity, Hidden Markov Model