[Studies in Computational Intelligence 481] Artur Babiarz, Robert Bieda, Karol Jędrasiak, Aleksander Nawrat (auth.), Aleksander Nawrat, Zygmunt Kuś (eds.) - Vision Based Systemsfor UAV Applications (2013, Sprin
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
234 H. Josiński et al.<br />
However, <strong>in</strong> a development phase of a human identification system gait databases,<br />
such as CASIA, CMU MoBo, NIST/USF, Soton, UMD, can be also used as data<br />
sources for classification purposes.<br />
Motion data lie <strong>in</strong> high-dimensional space. It is stated <strong>in</strong> [3] that many<br />
classifiers perform poorly <strong>in</strong> high-dimensional spaces given a small number of<br />
tra<strong>in</strong><strong>in</strong>g samples. Thus, feature extraction or dimensionality reduction is an<br />
attempt to transform a high-dimensional data <strong>in</strong>to a low-dimensional equivalent<br />
representation while reta<strong>in</strong><strong>in</strong>g most of the <strong>in</strong>formation regard<strong>in</strong>g the underly<strong>in</strong>g<br />
structure or the actual physical phenomenon [4].<br />
The research was aimed at verification of the hidden Markov model (HMM)<br />
accuracy as a gait video data classifier as well as at comparative analysis of<br />
quality of l<strong>in</strong>ear and non-l<strong>in</strong>ear methods of dimensionality reduction with regard<br />
to classification of reduced gait data.<br />
Generally, gait-based identification approaches can be divided <strong>in</strong>to two<br />
categories: model-free and model-based. The former category can be split <strong>in</strong>to<br />
approaches based on a mov<strong>in</strong>g shape and those which use <strong>in</strong>tegrate shape and<br />
motion with<strong>in</strong> the description [2]. In the first example of the model-free approach<br />
silhouettes of walk<strong>in</strong>g human be<strong>in</strong>gs were extracted from <strong>in</strong>dividual frames us<strong>in</strong>g<br />
background subtraction, their morphological skeletons were computed and the<br />
modified <strong>in</strong>dependent component analysis (MICA) was proposed to project the<br />
orig<strong>in</strong>al gait features from a high-dimensional measurement space to a lowerdimensional<br />
eigenspace. Subsequently, the L2 norm was used to measure the<br />
similarity between transformed gaits [5]. The pr<strong>in</strong>cipal components analysis<br />
(PCA) was also used for the purpose of human recognition [6], [7]. In the latter<br />
case three types of motion: slow walk<strong>in</strong>g, fast walk<strong>in</strong>g and walk<strong>in</strong>g with a ball<br />
were taken <strong>in</strong>to consideration. In [8] the recognition process was based on<br />
temporal correlation of silhouettes, whereas a spatio-temporal gait representation,<br />
called gait energy image (GEI), was proposed <strong>in</strong> [9]. The application of the<br />
Procrustes shape analysis method and the Procrustes distance measure <strong>in</strong> gait<br />
signature extraction and classification was shown <strong>in</strong> [10]. Numerous studies, <strong>in</strong>ter<br />
alia [11], [12], present frameworks developed for recognition of walk<strong>in</strong>g persons<br />
based on the dynamic time warp<strong>in</strong>g technique (DTW).<br />
The model-based approaches use <strong>in</strong>formation about the gait, determ<strong>in</strong>ed either<br />
by known structure or by model<strong>in</strong>g [2]. The Acclaim ASF/AMC format is often<br />
applied as the skeleton model of the observed walk<strong>in</strong>g person. Numerous methods<br />
aim to estimate the model directly from two-dimensional images. In [13] the<br />
particle swarm optimization algorithm (PSO) is used to shift the particles toward<br />
more promis<strong>in</strong>g configurations of the human model. In [14] 2D motion sequences<br />
taken from different viewpo<strong>in</strong>ts are approximated by the Fourier expansion. Next,<br />
the PCA is used to construct the 3D l<strong>in</strong>ear model. Coefficients derived from<br />
project<strong>in</strong>g 2D Fourier representation onto the 3D model form a gait signature.<br />
Another set of features used for human identification is extracted from spatial<br />
trajectories of selected body po<strong>in</strong>ts of a walk<strong>in</strong>g person (root of the skeleton, head,<br />
hands, and feet), named as gait paths [15].