Three-dimensional dynamic gesture recognition method based on deep learning

Abstract

The invention provides a three-dimensional dynamic gesture recognition method based on deep learning and stereoscopic vision, which has low cost, high precision, high speed and good timeliness. The deep learning method can be used to quickly extract feature points of a gesture area, meeting the requirement on real-time dynamic three-dimensional reconstruction, and solving the problem that the stereoscopic vision technique has matching difficulty. The method using the stereoscopic vision technique has lower cost than structured light scheme, depth camera scheme and other schemes. By providing sparse reconstruction for feature points, calculation quantity is greatly decreased at the premise of no precision loss. Trajectory recognition speed and precision are increased for a feature point tracing method; a trajectory defining scheme based on feature points also decreases trajectory recognition and matching difficulties.

Description

Technical field: [0001] The invention belongs to the technical field of image processing, in particular to a method and device for optical three-dimensional reconstruction and recognition of dynamic gestures. Background technique: [0002] As an important part of computer vision, gesture recognition is a bridge between the real world and the virtual digital world, and is widely used in many industries such as virtual reality, augmented reality, human-computer interaction, digital entertainment, and e-commerce. The traditional gesture recognition method for human-computer interaction is mainly based on touch interaction devices (mouse, touch screen). This type of interaction method cannot completely simulate gesture actions, and can only perform two-dimensional action recognition. Optical measurement is generally used to obtain three-dimensional gesture information. method. [0003] Due to the advantages of non-contact, high precision and fast speed, optical three-dimensiona...

AI Technical Summary

Problems solved by technology

This method uses a stereo vision system to learn and extract the key feature points of the gestures on the images collected by the left and right cameras through deep learning methods to solve the problem of "difficult matching". Coordinates, and finally track the three-dimensional coordinate trajectory of the key points of the multi-frame acquisition image, compare it with the pre-defined trajectory model, and judge the meaning of the dynamic gesture

Images