GESTURE RECOGNITION FOR HUMAN-MACHINE INTERACTION IN AVIATION

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Yıldız Teknik Üniversitesi, Fen Bilimleri Enstitüsü, AVİYONİK MÜHENDİSLİĞİ ANABİLİM DALI, Türkiye

Tezin Onay Tarihi: 2022

Tezin Dili: İngilizce

Öğrenci: TUĞBA ZEYBEK

Danışman: Ufuk Sakarya

Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu

Özet:

With the developing technology, human-machine interaction (HMI) is very significant in aviation, as in every area. Hand gestures are accepted in HMI applications. Because the human hand is quite adept at controlling a device. Gesture recognition problems in HMI also have some problems due to their nature. Not every human performs the same movement in space in the same way. Even the same person cannot perform the same movement in the exact same manner. This complicates the recognition problem. Due to large in-class differences between users, the efficiency of features derived from different sensors is compared in user-independent situations. The spatio-temporal nature of gesture signals is considered for recognition and a simple statistical feature-based approach is presented that works on transformed signals. With this approach, it is time-independent and allows the transition to the energy domain. With the wavelet-based statistical feature method, the size of the new data increases considerably. It becomes difficult to analyze high-dimensional data and to find meaningful data in the data set. Linear dimension reduction techniques are applied to solve the new problem. The scenario created in the motion classes is classified after the combination of dimension reduction and classification techniques. These results are linked to the flight simulator tool. In this thesis study, in cases where the pilots are insufficient for the increasing number of UAVs, the dominant set clustering method is recommended in order to provide UAV control with less trained pilots. This method is flexible as it can be used in both civil and military fields. In critical situations, more than one intermediately trained pilot may act like a well-trained pilot. In other cases, already moderately trained pilots will suffice. Keywords: Human-Machine Interaction, gesture recognition, spatio-temporal, Support Vector Machine, dominant set clustering