Hybrid Correlation-Neural Network Synergy for Gait Signal Classification - Advances in Heuristic Signal Processing and Applications - page 260

identification of gait pattern. However, as compared to a young-elderly differentia-
tion method, development of a disease identification scheme is a more complicated
problem, as it is a multi-class (more than two) classification problem.
The present chapter presents the study on the development of an automated gait
identification tool which can automatically determine whether or not the subject un-
der consideration is a healthy one, and if not, then whether the source of neurological
disorder in the pathological subject is due to Parkinson's disease (PD), Huntington's
disease (HD), or Amyotrophic Lateral Sclerosis (ALS). Thus, the overall purpose of
the proposed method is to predict whether an unknown subject under consideration
is healthy or suffering from one of the three major neurological diseases.
The basic problem of designing such gait identification tools can be divided into
two subworks:

(i) Suitable feature extraction from input gait signals;
(ii) Designing a suitable classification algorithm to utilize those extracted features.

Feature extraction can be conventionally carried out from input signals by us-
ing various mathematical tools like statistical methods, Fourier transform, wavelet
transform based methods, etc. This chapter attempts to develop efficient feature ex-
traction algorithms employing correlation techniques, instead of the above men-
tioned methods. The chapter explores cross-correlation as a potential tool for fea-
ture extraction of gait signals. Both time and frequency domain based features from
correlations are analyzed to develop powerful gait signal classification algorithms.
The cross-correlation technique has so far been conveniently utilized in several en-
gineering fields, e.g., in instrumentation, robotics, and remote sensing applications.
The cross-correlation technique has also been successfully used in sonar and radar
systems for range and position detection. The chapter also aims at investigating the
usefulness of employing Elman's recurrent neural network (ERNN) based classi-
fiers, on the basis of the features extracted employing cross-correlation techniques.
ERNN has found successful applications in the domains of function approximation,
prediction, and pattern recognition. Hence, the goal of this chapter is to develop
computer-based highly reliable automatic classification algorithms which can effec-
tively classify gait signals, utilizing cross-correlation based feature extraction and
neural network based classification techniques.

12.2 The Acquisition of Gait Signals

Several researchers have, over a period of time, employed different methodologies
for the analysis of gait signals. Most of these schemes employ different types of
acquisition procedures for recording various signals that are in some way related
to gait and posture of a human body. Subsequently, a variety of mathematical tech-
niques have been utilized for extracting meaningful features from these acquired
signals. Many of these methodologies are based on the recording of [ 11 ]:

• Step frequency or cadence,