http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Liao Jianwen,Meng Wei,Liu Zhifa,Cui Jinqiang 대한전기학회 2022 Journal of Electrical Engineering & Technology Vol.17 No.4
This paper presents a lightweight convolutional neural network architecture based hand gesture recognition for micro aerial vehicle (MAV) applications, which is able to detect hand gestures accurately and quickly convert into commands to control the MAV’s fl ight. In order to improve detection accuracy, a new activation function, named as adjustable-arctangent linear units (ALU) is proposed combined with linear stretching anchors (LSA) for accurate hand gesture detection. To achieve high detection speed, a lightweight neural network architecture (LNNA) is proposed and well trained. A large number of experiments show that by using LSA and ALU, the proposed solution can achieve the highest detection accuracy performance compared with commonly used methods. Moreover, compared with currently used YOLOv3, LNNA can improve the detecting speed by about 48.9% on 2080Ti GPU, and increase the speed to 11.67 frames per second on NVIDIA Jetson TX2. Flight experiments on MAVs also proved the eff ectiveness of the proposed hand gesture detection method.