An embedded system-based hand-gesture recognition for human-drone interaction
Khadidja Belattar, Abdelhak Mehadjbia, Abdelkarim Bala, Ahmed Kechida · International Journal of Embedded Systems · 2022
Using hand gestures is a natural mode of communication in human-robot interaction. This article presents an embedded command system for human-drone interaction. The main significance of the system is the improvement of human-drone interaction. Three contributions are proposed. The first one is the construction of a static hand-gesture dataset of three basic classes namely: 'start recording', 'take a photo' and 'stop recording'. The second contribution consists of the investigation of different transfer learning-based deep object detectors in both laptop and embedded computer platforms. The third contribution is an embedded hand-gesture recognition system, which allows users interaction, in real-time, with the quad-rotor drone using hand-gestures. The experimental results demonstrate the superiority of the YOLOv5-small model in the embedded recognition system. It yielded 98.32% mean average precision, 93.91% recall, 0.0025 loss and 0.92 recognition speed. Overall, the obtained results are promising and could be exploited in real-time object tracking applications.