Extraction of Geometric and Prosodic Features from Human-Gait-Speech Data for Behavioural Pattern Detection: Part II

This part of the research paper emphasizes on the detection of behavioural pattern from the extracted prosodic and geometrical features using human-gait-speech data. The clusters of these above-extracted features are mapped for the detection of behavioural pattern using soft-computing technique and c-means clustering method. Here, only four features of human-gait and four features of human-speech pattern are used for the formation of clusters. These clusters are mapped to close vicinity with minimum distance measurement, which in return is helpful for the proper classification and decision process, with a positive outcome for the detection of behavioural pattern. The mapping has been done with proper mathematical analysis over each feature of human-gait-speech pattern. The four prosodic features (extracted from human-speech pattern) are speech duration, speech rate, pitch and speech momentum, whereas the four geometrical features (extracted from human-gait pattern) are step length, walking speed, energy or effort and gait momentum, which are clustered. Here, five different natural languages (Hindi, Bengali, Oriya, Chhattisgarhi and English) have been employed for the completion of this part of research, when the subject is talking while walking. The classification process is being carried out with the help of a human-gait-speech-model (HGSM), using Baye’s theorem and support vector machine of artificial neural network. The mapping process has been carried out using adaptive-unidirectional-associative-memory (AUTAM) technique with an acceptable limit. The decision process for the detection of behavioural pattern has been done using revolutionary algorithm called genetic algorithm. Three behavioural patterns have been detected with three class-based moments: happy moments, normal moments and sad moments. An algorithm, called behavioural pattern detection algorithm using human-gait-speech pattern (BPDAHGSP), has been proposed. The complexity measures have been done, and the performance of the overall authentication system has been found very helpful for promoting global biometrical security system using nominal number of features.