Research On Dynamic Gesture Design And Recognition Method Based On RealSense

With the progress of science and technology,human-computer interaction technologies have made considerable development,which are becoming more humane and natural.Traditional human-computer interaction interfaces,such as command line interface and graphical user interface,have been unable to satisfy people's needs.The new generation of human-computer interaction is based on voice,touch and gesture.Compared with other parts of the body,hand is the most effective human-computer interaction tool.The hand movement is simple and intuitive,and the interaction can be completed without physically touching the machine.Therefore,gestures will become one of the hot research areas in human-computer interaction in the future.The research of gesture recognition mainly uses data glove and camera.Data glove which can get more accurate gesture data is expensive,inconvenient to wear,and is difficult to apply in large scale in practice.The camera needed for vision-based gesture recognition is more easily popularized.With the advent of 3D cameras such as RealSense and Kinect,gesture recognition based on vision has made greater progress,and has become one of the hottest research areas.The limitation of the existing gesture recognition researches is only focusing on the recognition of standard gestures in the gesture database which makes it impossible to perform real-time gesture recognition or apply in real scene.Aiming at the above problems,we first designed a set of basic dynamic gesture used in teaching to interact with teaching software.Second,using the basic data provided by the RealSense 3D camera,the continuous gesture segmentation algorithm based on the state automata is used to determine the beginning and end of the gestures so as to obtain the complete gesture data.According to different kinds of gestures,we choose features and construct feature vectors.In recognition part,we construct classification tree and combine DTW to realize dynamic gesture recognition.And the gesture recognition method is optimized for teaching scenes,which effectively reduces the recognizing error rate.Finally,a real-time gesture recognition system has been designed and implemented,which included gesture recording,gesture recognizing and application.These modules realized hand gesture recording and recognition respectively,and took place of some functions of mouse and keyboard by using gestures.In order to verify the validity of the gesture recognition system,this paper sets up a standard gesture database for seven basic dynamic gestures and identifies the gestures in the database.On this basis,a real time gesture recognition experiment is carried out.The later verification experiments show that the continuous gesture segmentation algorithm can accurately determine the beginning and end of the gestures,and divide the effective gestures completely.The recognition rate of gesture recognition algorithm reaches 96%,which has the characteristics of high accuracy and strong real-time.By combining gesture recognition system with classroom teaching,we can use gesture to manipulate multimedia devices.Teachers in the classroom could teach more flexibly and interesting.