Achenbach, Philipp ; Purdack, Dennis ; Wolf, Sebastian ; Müller, Philipp Niklas ; Tregel, Thomas ; Göbel, Stefan (2022)
Paper Beats Rock: Elaborating the Best Machine Learning Classifier for Hand Gesture Recognition.
8th Joint International Conference on Serious Games (JCSG'22). Weimar, Germany (22.-23.09.2022)
doi: 10.1007/978-3-031-15325-9_17
Konferenzveröffentlichung, Bibliographie
Kurzbeschreibung (Abstract)
More and more digital experiences, such as Serious Games, rely on gesture control as a means of natural human communication. Therefore we investigated the suitability of the Senso Glove: DK2 data glove and a Support Vector Machine for recognizing static hand gestures of the popular game Rock-Paper-Scissors in a previous work. Building on this, we now want to increase the scope of training and testing data and evaluate different kinds of Machine-Learning classifiers in addition to the Support Vector Machine. For this purpose, we ingested two different datasets, optimized them using grid search, and evaluated all user data in such a way that each user dataset was used individually for testing (leave-one-out) in order to obtain the most possible representative and user-independent result. Our results show that for a small number of gestures, Logistic Regression has the highest accuracy (97.6%) in predicting the results quickly. For a larger dataset, Random Forest achieves the highest accuracy (82.4%). Random Forest and Logistic Regression give very good results on both datasets (average 89.7%, both), but Logistic Regression is significantly faster overall. If the training and test data are not separated by user and are thus user-dependent, the results for both data sets improve to 99.2% and 99.5% with Support Vector Machine, respectively, and again Random Forest performs very well, with Logistic Regression showing small weaknesses here. In addition, we investigated how the accuracy of the classifiers performed when we gradually reduced the number of gestures from 25 to three in a dataset with 25 gestures and found that up to 11 gestures, a high accuracy of more than 94% could be achieved.
| Typ des Eintrags: | Konferenzveröffentlichung |
|---|---|
| Erschienen: | 2022 |
| Autor(en): | Achenbach, Philipp ; Purdack, Dennis ; Wolf, Sebastian ; Müller, Philipp Niklas ; Tregel, Thomas ; Göbel, Stefan |
| Art des Eintrags: | Bibliographie |
| Titel: | Paper Beats Rock: Elaborating the Best Machine Learning Classifier for Hand Gesture Recognition |
| Sprache: | Englisch |
| Publikationsjahr: | 14 August 2022 |
| Verlag: | Springer |
| Buchtitel: | Serious Games: Joint International Conference - JCSG 2022 |
| Reihe: | Lecture Notes in Computer Science |
| Band einer Reihe: | 13476 |
| Veranstaltungstitel: | 8th Joint International Conference on Serious Games (JCSG'22) |
| Veranstaltungsort: | Weimar, Germany |
| Veranstaltungsdatum: | 22.-23.09.2022 |
| DOI: | 10.1007/978-3-031-15325-9_17 |
| Kurzbeschreibung (Abstract): | More and more digital experiences, such as Serious Games, rely on gesture control as a means of natural human communication. Therefore we investigated the suitability of the Senso Glove: DK2 data glove and a Support Vector Machine for recognizing static hand gestures of the popular game Rock-Paper-Scissors in a previous work. Building on this, we now want to increase the scope of training and testing data and evaluate different kinds of Machine-Learning classifiers in addition to the Support Vector Machine. For this purpose, we ingested two different datasets, optimized them using grid search, and evaluated all user data in such a way that each user dataset was used individually for testing (leave-one-out) in order to obtain the most possible representative and user-independent result. Our results show that for a small number of gestures, Logistic Regression has the highest accuracy (97.6%) in predicting the results quickly. For a larger dataset, Random Forest achieves the highest accuracy (82.4%). Random Forest and Logistic Regression give very good results on both datasets (average 89.7%, both), but Logistic Regression is significantly faster overall. If the training and test data are not separated by user and are thus user-dependent, the results for both data sets improve to 99.2% and 99.5% with Support Vector Machine, respectively, and again Random Forest performs very well, with Logistic Regression showing small weaknesses here. In addition, we investigated how the accuracy of the classifiers performed when we gradually reduced the number of gestures from 25 to three in a dataset with 25 gestures and found that up to 11 gestures, a high accuracy of more than 94% could be achieved. |
| Fachbereich(e)/-gebiet(e): | 18 Fachbereich Elektrotechnik und Informationstechnik 18 Fachbereich Elektrotechnik und Informationstechnik > Serious Games |
| Hinterlegungsdatum: | 24 Jan 2023 10:52 |
| Letzte Änderung: | 07 Mär 2023 13:20 |
| PPN: | 505589540 |
| Export: | |
| Suche nach Titel in: | TUfind oder in Google |
 |
Frage zum Eintrag |
Optionen (nur für Redakteure)
 |
Redaktionelle Details anzeigen |
Drucken
Impressum