Human-AGV Interaction: Real-Time Gesture Detection Using Deep Learning

摘要

In this paper, we present a real-time human body gesture recognition for controlling Automated Guided Vehicle (AGV) in facility. Exploiting the breakthrough of deep convolutional networks in computers, we have developed a system that can detect the human gestures and give corresponding commands to the AGV according to different gestures. For avoiding interference of multiple operational targets in an image, we proposed a method to filter out the non-operator. In addition, we propose a human gesture interpreter with clear semantic information and build a new human gesture dataset with 8 gestures to train or fine-tune the deep neural networks for human gesture detection. In order to balance accuracy and response speed, we choose MobileNet-SSD as the detection network.