Method and system for dynamic gesture recognition

Abstract

The invention discloses a dynamic gesture recognition method. The method comprises the following steps: shooting the gesture by a 3D camera to obtain a group of hand motion videos with n frames of images; identifying m hand feature points in each frame image respectively, obtaining spatial coordinate information of each hand feature point in the corresponding image frame, and obtaining a two-dimensional vector m*3 of hand feature in one frame image; forming a three-dimensional vector set n*m*3 from a two-dimensional vector set of hand features in each frame image; steps S1-S3 are repeated, acquiring a plurality of three-dimensional vector sets; The obtained plurality of the three-dimensional vector sets are used as training samples, a gesture name is marked on the training samples, and model training is performed on the training samples to obtain model parameters corresponding to the gesture names, so that dynamic gesture recognition can be performed with high accuracy.

Description

technical field [0001] The invention belongs to the technical field of recognition in the field of communication and information, and in particular relates to a dynamic gesture recognition method and system. Background technique [0002] At present, with the rapid development of artificial intelligence technology, gestures have gradually become one of the important sources of device input for a variety of human-computer interaction or augmented reality tasks. Therefore, accurate gesture estimation is becoming more and more important, and in The field of computer vision has attracted increasing attention. [0003] In terms of hardware, with the development of cameras and display hardware levels, the representation of gestures has slowly developed from the initial 2D space to 3D space, and the spatial descriptiveness of shapes has been greatly enhanced, and it is getting closer to the real sensory understanding of humans. In terms of software, with the development and progres...

AI Technical Summary

Problems solved by technology

The disadvantage of this technical solution is that the high-level abstract features of each frame obtained by CNN convolution have no human intervention. In other words, the obtained abstract features are only learned by the network itself, and the network "thinks" that these features can be used as The recognition basis of different gestures does not necessarily conform to our cognition. If the background of the gesture is more complicated, or there are similar objects, then the features obtained by convolution will be different, and the corresponding gesture recognition will have a certain degree of accuracy. error, even misjudgment

Images