Research On Dynamic Gesture Recognition Based On Temporal And Spatial Features

In recent years,with the rapid development of science and technology,artificial intelligence,augmented reality and other emerging technologies have gradually entered people's vision,and gradually penetrated into all aspects of life.The research in the field of computer vision is booming under the influence of emerging technologies.As the main research field of computer vision,gesture recognition is not only widely used in all walks of life,but also deeply studied by more and more scholars.It has become a research hotspot in the field of computer vision and has broad application prospects.Dynamic gesture recognition is easily affected by light changes,complex environment,fast change and other factors,and has the problem of low recognition rate.Therefore,the related research based on dynamic gesture recognition has great research value.In this paper,the algorithm of dynamic gesture recognition is mainly studied for the problems of easy to be affected by light changes and complex environment.The research work of this paper is as follows:Aiming at the problem of large redundancy in dynamic gesture video sequence,this paper firstly extracts the key frames of video sequence based on the inter frame difference algorithm,and extracts the key frame image by using the change of pixel value of each frame.Secondly,the extracted key frames are extracted by a multi-scale Retinex with chromaticity preservation,MSRCP algorithm for low illumination image enhancement can effectively solve the problems of unclear local details and poor image visibility,which is convenient for subsequent feature extraction.Then,the key points of the hand are extracted with Open Pose algorithm.Finally,XGBoost(extreme gradient boosting)integration algorithm is used to realize the classification of dynamic gesture sequences.The complexity of dynamic gesture recognition is high,the amount of information in video sequence is large,and the motion features are obvious.Therefore,this paper uses dense optical flow optimization algorithm to extract the dynamic motion feature information of dynamic gesture.In this paper,we mainly choose farneback dense optical flow algorithm to extract the direction vector and size information of gesture motion change,and fuse the gesture coordinate features extracted earlier.Finally,we use xgboost algorithm to classify and recognize the fused features.In order to verify the effectiveness of the proposed algorithm,experiments are carried out on three open datasets,including Microsoft Kinect and Leap Motion datasets,UTD -MHAD datasets and ChaLearn LAP IsoGD datasets.The optimization algorithm is verified by UTD-MHAD datasets and ChaLearn LAP IsoGD datasets.Comparative experiments show that the proposed algorithm has a certain improvement in recognition rate compared with the excellent algorithms in recent years,which verifies the effectiveness of the proposed algorithm.