Realtime Recognition of Dynamic Hand Gestures in Practical Applications

Dynamic hand gesture acting as a semaphoric is practical and intuitive mid-air interface. Nowadays benefiting from the development of deep convolutional networks, recognition has already achieved high accuracy, however, when performing dynamic such gestures direction commands, some unintentional actions are easily misrecognized due to similarity poses. This hinders application cannot be solved by just improving accuracy applied algorithm on public datasets, thus it necessary study problems perspective human-computer interaction. In this article, two methods proposed avoid misrecognition introducing activation delay using asymmetric design. First temporal process decomposed redefined, then realtime system built through two-dimensional neural network. order investigate influence design performance, user conducted experimental results show that can effectively misrecognition. The in article provide valuable guidance for researchers designing applications.