Back to results

Aristotle University Of Thessaloniki (AUTH)

Ευφυής λήψη και επεξεργασία βιοσημάτων για την αυτοματοποιημένη αναγνώριση χειρομορφών νοηματικής γλώσσας

Abstract

dc:description

The main research field inspected in this doctorate dissertation focuses on the automated Greek Sign Language (GSL) recognition using data from surface electromyogram and accelerometer. In this context, the requirements of the acquisition system, such as the type, number and position of the electromyogram electrodes as well as the proposed position of the triaxial accelerometer are presented. The optimum topology consisting of five-channel surface electromyogram, capturing the wrist and finger motions and three-channel acceleration, monitoring the position of the hand in the 3D-space is reported in detail. An extended set of time-frequency features proposed in the literature are extracted from the acquired signals and evaluated for their efficiency in correctly classifying the isolated signs. As presented in this dissertation, the time-frequency features are inadequate for a medium sized sign lexicon, implying the necessity of employing more advanced signal processing techniques. To this end, advanced methods such as the entropy and the empirical mode decomposition are employed to provide with high recognition hit-ratios, along with feature vector dimension reduction, towards fast and reliable automated GSL recognition. In order to discriminate the differences in the individual way of signing from those regarding the sign gestures and therefore enhance the GSL recognition results, the signer’s level of deafness is employed in the feature extraction process. Through a novel hybrid adaptive procedure, a genetic algorithm weights the intrinsic modes of the acquired signals, before the sample entropy calculation. The resulting feature set, namely weighted intrinsic mode entropy, aims at minimizing the feature space dimension and increasing the sign classification accuracy discarding the intra-signer differences. The biggest contribution of this approach lies in the construction of a signer-independent GSL classification system towards the enhancement of its performance (>90% for 60 signs). To ensure the credibility of the presented results, the vocabulary complexity in terms of the similar movements involved in each hand articulation was examined. A novel GSL modeling approach on the kinetic level is proposed enabling the comparison of the signs through the calculation of the degree of similarity. The results imply sufficient confusion in the selected vocabulary for the evaluation of the robustness of the proposed methods. Another novelty in this dissertation is the use of the acquired data for the signer and the signer's level-of-deafness recognition. The findings justify the employment of the level of deafness in the feature space linking the individual way of signing to the signer identity and the level of deafness. This doctorate dissertation via the proposed data acquisition and advanced data processing techniques paves the way for more efficient automated GSL recognition. The proposed analyses can be useful in many gesture recognition applications, such as human-computer interfaces and robotics.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2010

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Κοσμίδου, Βασιλική
  • Kosmidou, Vasiliki

Subjects

dc:subject × 12

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/28933

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Κοσμίδου, Βασιλική; Kosmidou, Vasiliki. Ευφυής λήψη και επεξεργασία βιοσημάτων για την αυτοματοποιημένη αναγνώριση χειρομορφών νοηματικής γλώσσας. Aristotle University Of Thessaloniki (AUTH), 2010. http://hdl.handle.net/10442/hedi/28933