Characteristic covariance matrix-based dynamic gesture recognition method

Abstract

The invention discloses a characteristic covariance matrix-based dynamic gesture recognition method. According to the method, a uniform framework is provided for two-dimensional dynamic gestures and three-dimensional dynamic gestures; for recognition of the two-dimensional dynamic gestures, only FAST angular points of hand areas in three continuous frames are accurately tracked, and the whole hands do not need to be accurately tracked, so that the tracking requirement is greatly reduced and the realization is easier; for recognition of the three-dimensional dynamic gestures, three-dimensionalskeleton information of hands is obtained in real time by using an SDK of a RealSense F200, and obtaining a three-dimensional coordinate of hand joint points in each frame so as to robustly track thegestures; and through logarithmic covariance matrix descriptors, compact expressions are provided for the whole gesture sequences, so that the SVM model training efficiency and efficiencies of gesturerecognition systems are greatly improved.

Description

technical field [0001] The invention relates to the technical field of image processing and analysis, in particular to a dynamic gesture recognition method based on a feature covariance matrix. Background technique [0002] Vision-based gesture recognition plays an extremely important role in human-computer interaction, robot control, virtual reality and other fields, so it has received extensive attention in the field of computer vision. Generally speaking, gestures can be divided into two types: static gestures and dynamic gestures. Static gesture recognition mainly recognizes hand gestures from a single image, so it is largely limited to practical application scenarios. Compared with static gestures, dynamic gestures include the movement process of the hand and can convey more information. In addition, dynamic gestures are generally collected in a moving environment, making dynamic gesture recognition more widely used in application scenarios. [0003] Dynamic gestures...

AI Technical Summary

Problems solved by technology

However, so far, few studies have been done on dynamic gesture recognition based on hand joint features.

[0006] It can be seen that the existing algorithms in the field of dynamic gesture recognition still have many shortcomings, especially the hand tracking, model training and template matching efficiency in two-dimensional dynamic gesture recognition.

Images