Performance evaluation of gait recognition using the largest inertial sensor-based gait database

Abstract

This paper presents the largest inertial sensor-based gait database in the world and its application to a statistically reliable performance evaluation for gait-based recognition problem. Whereas existing gait databases include at most a hundred subjects, we construct a much larger gait database for both accelerometer and gyroscope which includes 736 subjects (382 males and 354 females) with ages ranging from 2 to 78 years. Because a sufficiently large number of subjects for each gender and age group are included in this database, we can analyze the dependence of gait recognition performance on gender or age groups. The results with the latest existing recognition method provide several novel insights, such as the trade-off of gait recognition performance among age groups derived from the maturity of walking ability and physical strength. Moreover, the evaluation for the recognition performance improvement with a larger number of subjects was reliably confirmed in the experiments. As for sensor data type, acceleration is better than angular velocity for gait recognition performance. Compared to unnormalized distance (such as Euclidean distance), normalized distance (such as normalized cross correlation-based distance) works significantly better for angular velocity.