Research on gesture recognition based on YOLOv5

Abstract

In recent years, with the rapid development of 5G, artificial intelligence, and Internet of Things technologies, smart digital home have entered thousands of households, which made our lives colorful. As a kind of human-computer interaction (HCI), gesture has the characteristics of strong information expression ability and transmission function, non-contact, and simple and easy to understand, which has been widely used in robot control, virtual reality, intelligent driving, and somatosensory games. The initial interaction method of gesture was through external hardware such as digital gloves, biological myoelectricity, and Kinect depth equipment, and then gradually developed into a method based on computer vision algorithms. The rapid development of deep learning in recent years has greatly improved the performance of image recognition and provided a new direction for gesture recognition. The deep learning method based on convolutional neural network can automatically extract image features, and is less affected by image background factors, which greatly improves the recognition accuracy. With the development of target detection technology, some algorithms have been increasing the depth of the network to obtain higher recognition accuracy, which will lead to excessive computing resource and high hardware requirements. Complex neural networks slow down the model recognition rate and are difficult to apply to resource-constrained mobile and embedded devices. This paper introduces the overall structure of the YOLOv5 network. The model is trained by YOLOv5s network and validated by real gesture. Finally, a gesture recognition software demo is made by PyQT5. The experimental results show that gesture recognition based on YOLOv5 can achieve better recognition accuracy.