Gesture Recognition And Its Application In Human-computer Interaction

Gesture recognition,as one of the most natural interactive methods,has high application value in human-computer interaction.Gesture recognition based on monocular RGB camera has the advantages of simple equipment and low cost.Therefore,it has a good prospect to enhance the robustness and rapidity of monocular RGB camera gesture recognition method for human-computer interaction.This paper aims to study the gesture detection method,gesture tracking method,dynamic gesture recognition method and fingertip recognition method under the monocular RGB camera,and improve the relevant methods according to the characteristics of gestures and apply them to the human-computer interaction system.This paper is mainly divided into the following four aspects:1)Gesture detection: Firstly,the characteristics and shortcomings of commonly used skin color modeling methods,machine learning detection methods and deep learning detection methods in gesture detection are analyzed.Then,combined with the characteristics of gesture in human-computer interaction,an improved multi-scale gesture detection method is proposed,which combines Mixed Gauss skin color modeling,Bayesian correction and extreme learning machine.Finally,the gesture data is built by ourselves.The comparative experiments on GEST_SCUT are carried out to verify the effectiveness of the improved detection method.2)Gesture tracking: Firstly,it analyzes the characteristics and limitations of common moving target detection methods in motion gesture detection,and the characteristics and limitations of common machine learning tracking methods in gesture tracking.Then,by designing the confidence model and introducing scale filters,an improved kernel correlation filter is proposed.Finally,the comparative experiments on GEST_SCUT are carried out to verify the effectiveness of the kernel correlation filter.3)Dynamic gesture recognition and fingertip point recognition: Firstly,the joint point of the gesture is generated by the gesture convolution machine,then the cosine direction features of the five fingertip points are extracted for the dynamic gesture,and a KNN-DW-DTW matching method is proposed to identify the dynamic gestures.Then the validity of the method is verified on the self-built data set GEST_SCUT.Finally,the fingertip point recognition method based on convex hull detection is analyzed,as well as the characteristics and shortcomings of the method are analyzed.An improved method combining convex hull detection,centroid distance and curvature detection is proposed.4)Human-machine interaction in aerial handwriting: Firstly,the functional requirements and program framework of the software system are introduced.Then,according to the requirement analysis of each functional module of the software,the corresponding design scheme is given.Finally,the results of software data acquisition,data annotation and aerial handwriting are given.