基于视觉的动态手势识别技术综述

Review of Research on Vision-based Dynamic Gesture Recognition

摘要:

动态手势识别是计算机视觉领域较为热门的任务之一，受到了研究者的广泛关注；动态手势识别技术在自动驾驶、虚拟现实和人机交互等诸多领域展现出很高的应用潜力；手势是在虚拟空间中与其他人交换信息、指导机器人在恶劣环境中执行特定任务或与计算机交互的一种直观而理想的方式；调研归纳了一些常用的动态手势数据集，对动态手势数据集的模态、数据量、应用场景进行了总结与分析；从使用方法的网络类别出发，综述了基于视觉的动态手势识别技术研究进展，重点介绍归纳了基于深度学习的方法，对基于卷积神经网络、循环神经网络以及图神经网络的方法进行了整理总结与性能比较；最后对基于视觉的动态手势识别的研究方向进行了展望。

Abstract:

Dynamic gesture recognition is one of the most popular tasks in the field of computer vision, which has been widely concerned by researchers. Dynamic gesture recognition technology has shown high application potential in many fields such as automatic driving, virtual reality and human-computer interaction. Gestures are an intuitive and ideal way to exchange information with others in a virtual space, to direct a robot to perform a specific task in a hostile environment, or to interact with a computer; Some commonly used dynamic gesture data sets are investigated and summarized, and the modes, data volume and application scenarios of dynamic gesture data sets are summarized and analyzed. Starting from the types of networks used, this paper summarizes the research progress of vision-based dynamic gesture recognition technology, focuses on introducing and concluding the methods based on deep learning, and summarizes and compares the methods based on convolutional neural network, recurrent neural network and graph neural network. Finally, the research direction of dynamic gesture recognition based on vision is prospected.