BACKGROUND: Gait recognition is an emerging biometric technology applied to the mobile environment. With built-in accelerometers, wearable devices are used to recognize user identity according to gait periodic pattern, which shows strong stability and uniqueness property.
OBJECTIVE: The purpose of this study is to build analyzing models to find the change of gait normal and pathological function based on gait features.
METHODS: This work relies on gait recognition methods. In this paper, the performance of different hybrid filter methods is compared by combining four classical filtering methods. The influence of the abnormal pattern of gait cycle is estimated by standard deviation. The effectiveness of feature matching methods is evaluated by six classical distance discrimination function.
RESULTS: The results highlight the stability and invariance of gait periodic pattern. For analyzing models, the best recognition rate is 96.67% with the combination of MF hybrid filter and Correlation distance function in the small sample, and minimal time consumption is 0.038 s. The effectiveness of analyzing models is further analyzed for different practical applications.
CONCLUSIONS: This study provides evidence for future scientific teams to make decisions on selecting filter methods and discrimination functions which can more efficiently extract gait features and suggest ways to analyze clinical gait pattern.