Eintrag in der Universitätsbibliographie der TU Chemnitz
Volltext zugänglich unter
URN: urn:nbn:de:bsz:ch1-qucosa2-1019892
Dadgar, Amin
Brunnett, Guido ; Kowerko, Danny (Gutachter)
Application-Independent Recognition of 3D Hand Gestures using Synthetic Data
Kurzfassung in englisch
There are numerous ways in which we could emphasize the profound role of our hands in our daily lives. We utilize our hands to point to a person or an object. We also use them to convey information about space, shapes, the number of objects, and the temporal characteristics of their movements \pprhyperref{Rautaray2012}. Our hands allow us to interact, relentlessly, with objects in a variety of ways including moving, modifying, and transforming them, as well as performing more precise activities in operating rooms, airplanes, laboratories, and factories. In addition, hand gestures are an efficient way for people to communicate and connect. To emphasize the degree of gesticulation in human conversation, it is instructive to note that people often continue to gesticulate even when speaking on the phone \pprhyperref{Garg2009}. All of these unconscious gesticulations are carried out to indicate thoughts such as 'stop'; 'come closer'; 'no' as we speak. We also employ them consciously, in sign language, to communicate with the deaf \pprhyperref{Starner1995}. Additionally, according to Luhmann \pprhyperref{Luhmann1992}, one cannot think without writing (Ohne zu schreiben, kann man nicht denken''). Could we even ask the following question: For an efficient thinking process, do the hands play a pivotal role, since the hands are our primary writing tool'? Based on all these versatile natural roles and wide ranges of influence from tangible to abstract activities that the hands have, it is clear that a system that integrates hand gestures and postures into its input set could benefit a wide range of industrial and academic sectors. In this doctoral dissertation, we advance the state of the art of hand gesture recognition systems and related fields within the \textit{vision-based} class of technologies on some of the existing crucial problems. In short, in \chphyperref{PhDThesis-ChpHierarDB} we propose a novel data structure based on hand kinematics to efficiently characterize the postures in three dimensions. Then, to model the temporal relationships between these postures (to form dynamic gestures), in \chphyperref{PhDThesis-ChpGestology} we propose a novel Markovian method based on computer vision techniques. Finally, to extract two-dimensional features from images, in \chphyperref{PhDThesis-ChpSaneNet} we employ and improve machine learning techniques. Specifically, we pace through top-down and bottom-up processes and combine them in the \textit{analysis-by-synthesis} \pprhyperref{Yuille2006} by introducing our novel methods in three phases: 1) by exploiting the kinematics and anatomy of the hands in pose space (top-down) to generate synthetic postures, 2) by exploiting the Dynamic Bayesian Network (DBN) \pprhyperref{Hoeting1999} in pose space (top-down) to formulate the posture movements into gestures, and 3) by employing the Artificial Neural Network (ANN) family \pprhyperref{Bengio2009} in the 2D image space (bottom-up) to train the convolutional neural networks \pprhyperref{Krizhevsky2012, Lecun2015, Moon2022} with synthetic images to address key challenges such as hand detection/tracking and segmentation in real images.
| Universität: | Technische Universität Chemnitz | |
| Institut: | Kompetenzzentrum Virtual Humans | |
| Fakultät: | Fakultät für Informatik | |
| Dokumentart: | Dissertation | |
| Betreuer: | Brunnett, Guido | |
| DOI: | doi:10.60687/2026-0035 | |
| SWD-Schlagwörter: | Computergrafik , Computervision , Maschinelles Lernen , Erkennung , Segmentierung | |
| Freie Schlagwörter (Englisch): | Computer Graphics , Computer Vision , Machine Learning , Statistical Temporal Modeling , Static Hand Postures , Dynamic Hand Gestures , Hierarchical Hand Data Structure , Detection , Segmentation , Synthetic Images , Fingers 1D PoseDescriptor | |
| DDC-Sachgruppe: | 006.37 | |
| Sprache: | englisch | |
| Tag der mündlichen Prüfung | 16.01.2026 | |
| OA-Lizenz | CC BY 4.0 |
