低コストのディープニューラルネットワークベース家電操作用リアルタイム手振り認識システム

抄録

With recent advancements in deep learning, the performance of hand gesture recognition (HGR) models has greatly increased, enabling reasonably accurate hand gesture detection with real time speed on modern desktop computer systems. Recently, there has been a growing interest in developing smaller and faster deep neural network (DNN) architectures suited for embedded devices. In this paper, we present a new DNN based end-to-end real-time control for home appliances. In purpose to gain reliable accuracy and execution speed, we proposed a new architecture by replacing the backbone of YOLOv3 and refining its head. The proposed network's performance showed advantages compared to popular object detection models. The model was incorporated into a HGR system for home appliance controls. The system provides a user friendly interface to register the correspondence between infrared signals and gestures. The proposed system takes images from a camera as input, recognizes gestures and produces corresponding infrared signals as output, providing a low cost system that can be used to control various home appliances under a complex environment. The code, trained tensor lite models and other software used in this research are available at https://github.com/appleyuta/RaspberryPi-HGR-System.