Research On Dynamic Gesture Instruction Recognition Strategy

Gesture recognition is one of the research hotspots at home and abroad.As an application technology of machine vision and human-computer interaction,gesture recognition is frequently used in many fields,For example: sign language recognition,remote sensing control of robotic arm,smart home applications,etc.At present,the research in this direction is limited by technology and environmental impact.There are still many problems to be solved and gesture instruction recognition methods are mostly based on functional gloves or using cameras with depth information to collect information,which requires strict conditions for application environment.This paper studies the technology of static and dynamic fusion gesture instruction recognition based on computer vision.It mainly uses a combination tracking algorithm based on skin color extraction and the Mean Shift tracking algorithm and the centroid and category features of gestures to implement tracking.Dynamic gesture instruction tracking and recognition has good robustness.Using a two-layer SVM training model reduces the dependence on the experimental environment.The main work is as follows:1)Customize the basic static gesture in 5,collect gesture with ordinary camera,and build up its own static gesture database,Using skin color segmentation to extract gestures,using region growing algorithm to eliminate interference areas,and extracting HOG features from target images.Through SVM classification,the extracted HOG features are multi-classified,and the training model of static gesture is obtained.Ten kinds of dynamic instructions are customized and recorded by a laptop camera.Dynamic gesture features are acquired by using static gesture training model and the position of each single frame gesture.Dynamic gesture model is acquired by using SVM trainer for the second time.2)Skin color models include RGB color space,HSV color space,YCr Cb color space,etc.Because skin color extraction in RGB space is influenced by chroma and brightness,The brightness V in HSV color space indicates the degree of brightness of the color.For light source color,the brightness value is related to the brightness of the luminous body and is sensitive to light.In YCr Cb space,Y represents brightness information,while Cb and Cr components are not affected by brightness,so Y,Cb and Cr are separated and skin color segmentation is carried out.3)Feature extraction is divided into static feature extraction and dynamic feature extraction.Static feature extraction mainly extracts HOG features based on the skin colorextraction results of two-dimensional gesture images.Dynamic gesture feature extraction is based on the static model to get the static gesture sequence of dynamic instructions whose sequence is obtained by normalizing the static gesture sequence with the centroid coordinates of each frame and the trajectory angle between adjacent frames obtained by Region growth.4)Combined with the principle of mean shift tracking algorithm and the contour and centroid of gesture obtained by region growing algorithm,a combined gesture tracking algorithm is constructed.The region growing algorithm can segment the gesture part in the binary image after the skin color processing and calculate the centroid of the contour at the frame.Therefore,this paper will take mean The shift tracking algorithm is combined with the gesture centroid and other information in this paper to design a combined tracking algorithm,which realizes the real-time accurate tracking of hand movement,and makes up for the low tracking accuracy of a single tracking algorithm.5)Experiment of Gesture Instruction Recognition.Static gesture experiments are carried out with a laptop camera to recognize different gestures made by different people.The experimental results show that the static model has good robustness to different people's non-action and meets the experimental requirements.Different gesture instructions made by different people are tested,and different experimental results are given according to the gesture instructions.The test results show that the recognition algorithm has good recognition efficiency,excellent robustness and real-time performance to meet the actual requirements.Finally,conclusions of this paper were drawn based on above experiments and pros and cons of proposed methods were analyzed with some suggestions of potential improvements.