Content area
Abstract
We propose a gait recognition system that incorporates both static and dynamic features for personal identification using both walking video and ground contact information (GAITRite™ instrumented walkway). Given the challenges of creating a robust gait recognition system, the aims of this study were: (1) to determine the effect of gait velocity and variable type (anthropometric vs. gait) on the matching performance of individual video and GAITRite™ measures; and (2) to determine the effect of gait velocity on multivariate gait matching performance.
To address the aims, both video and GAITRite™ data were collected from 94 healthy adult participants as they walked along a level pathway. Each subject completed five walking trials at three different gait velocities: self-selected velocity, self-selected velocity + 20%, and self-selected velocity −20%. A range of static and dynamic variables were then extracted from the data using techniques developed in preliminary studies. We evaluated the recognition performance of each variable by calculating Receiver Operating Characteristic (ROC) curves, and we compared the ROC measures to corresponding variability ratios (VR). The test variables were also combined to create multivariate gait recognition solutions. These solutions were tested to establish their recognition performance across different gait velocities and different variable groups (gait vs. anthropometric).
We observed a logarithmic relationship between AUC and VR that may be useful for predicting the matching performance of continuous variables. We also observed that the VR's for over half of the variables changed as a function of gait velocity. The results of the multivariate recognition analyses showed that there was no difference in recognition performance between the anthropometric and gait variables. Finally, the multivariate recognition solution containing toe in/out angle, foot length, and leading shank angle was shown to provide the best solution in the face of varying gait velocity.
