[An Efficient Human Identification Using Gait Analysis]

by

Literature Review

Introduction

Biometric systems are becoming increasingly important, as they provide more reliable and efficient means of identity verification. Human identification at a distance has recently gained enormous interest among computer vision researchers. Gait recognition aims essentially to address this problem by recognising people based on the way they walk. Human gait recognition works from the observation that an individual's walking style is unique and can be used for human identification. So as to recognize individual's walking characteristics, gait recognition includes visual cue extraction as well as classification. But the major issue here is the representation of the gait features in an efficient manner.

Feature extraction methods

Several feature extraction methods that are specific to biomechanics have been published. A novel approach for feature extraction from EMG measurements (Von Tscharner and Herzog, 2009) was introduced by von Tscharner (2010). The author used the property of the Wavelet transformation to give a representation of a signal both in frequency and in time. Thus, the features were capable of resolving events within the EMG signal in time, frequency and intensity. By averaging those features over multiple experiments, functional aspects of muscle activation could be determined. These so-called Wavelet features were applied in subsequent classification studies (von Tscharner, 2009: 41; von Tscharner and Goepfert, 2009: 166). An important idea in these studies was to use multi muscle Wavelet patterns for subsequent classification (von Tscharner and Goepfert, 2009: 28). These multi muscle patterns combined the information from several muscle groups. This approach allowed, for instance, employing timing relations between these muscle groups for classification.

A Wavelet feature extraction method was also used in a different study (Nyan et al., 2009: 191). The authors classified gait patterns from three classes: ascending stairs, descending stairs and level walking. A recognition rate of more than 97% was reported. As basis for the classification, the authors used data from accelerometers that were strapped to the shoulder of their subjects. Employing accelerometer measurements for such a task requires careful action by the researchers (Nigg and Boyer, 2009: 14). Especially the tightness of the strapping of the accelerometer has been shown to affect the measured signal amplitude. It has therefore yet to be demonstrated whether the information from this type of sensor is reliable enough to conduct more complex group classification tasks. Results were published using the kernel Principal Component Analysis for feature reduction in a study related to gait classification (Wu et al., 2009: 166). The feature reduction was applied to kinematic variables that were computed at specific time points of the gait cycle (e.g. heel-strike, toe-off). The computation of these kinematic variables can lead to an error amplification (e.g. when skin movement affects the marker positions) and requires additional assumptions (e.g. about the joint axis directions). Furthermore, a substantial amount of information is lost by using only specific time points of the measured time series. It has already been stated in the literature that the incorporation of time dependent patterns yields valuable information for gait analysis ...