Gait Analysis Using Shadow Motion

Abstract

Gait is considered as one of the biometric traits that does not require physical interaction with machines and can be performed at a distance from the computing device. However, majority of the gait recognition systems require the subjects to be monitored in constrained environment within the viewing field of the capturing device. Such systems may fail to recognize a few of the features when the interaction environment is changed or when the body occlusion occurs due to position variations, clothing or belongings. Moreover, the walking style of a user may vary when engaged in different activities such as listening to music, playing games, fast walking, etc. In this paper, we propose a new approach of human gait recognition using Shadow motion sensor, a full body sensor unit. The framework is able to identify users robustly despite changes in their appearances. The device uses a combination of accelerometer, gyroscope and magnetometer sensors for collecting gait features. The identification process is performed using a Random Forest based classification scheme by varying number of trees. A set of users comprising with 23 males and females have participated in the data collection and they have performed four different types of walks including, normal-walk, fastwalk, walking while listening to music and walking while watching video on mobile. An average accuracy of 87.68% has been recorded in all walk scenarios. Results reveal that the proposed study can be used as a stepping stone to design robust gait biometric systems with the help of contact less sensors.