Kinect-based Gesture Recognition And Its Application In Scenario-driven

At present,the human-computer interaction technology is one of the most rapidly growing technologies in the study of user interface and the researchers pay particular attention to it.It is a comprehensive discipline which is closely related to cognitive science,ergonomics,psychology and other subject areas.Gesture recognition has been taken seriously by many researchers as an important part of human-computer interaction.Especially in recent years,the research of gesture recognition interaction technology has become very active which is in accord with the man-machine communication habits with the emergence of Kinect of Microsoft.Gesture recognition research includes static gesture recognition and dynamic gesture recognition according to the classification of gestures action.In this topic,Kinect is provided by Microsoft as gestures collection devices,the algorithm of the static gesture recognition and dynamic gesture recognition is optimized respectively,then the test is completed in the virtual scene.First of all,a new algorithm of hand segmentation is put forward in order to make the hand segmentation more accurate.The algorithm can get the best segmentation threshold through calculating between-cluster variance of the torso area and hand area so as to extract to hand area.And calculate the maximum point in hand area points again to get palm point.Then the hand area is broken down into palm,fingers and arms which is based on describing the palm area adopted corresponding oval.Secondly,an algorithm of gesture recognition based on multi-threshold feature extraction is put forward for low accuracy rate while using single feature recognition.At first of all,the algorithm extracted the distance from fingertips point to the center of the palm,the distance from fingertips point to hand flat and the hand area,then static gestures are classified by Multi-Class Support Vector Machine.The algorithm verification is completed in the gesture database.Thirdly,a kind of algorithm which uses joint point credibility to measure joint point validity is put forward aiming at joint point acquired inaccurate in the course of dynamic gesture recognition.The algorithm can obtain more accurate joint point of dynamic hand gestures and complete dynamic gesture recognition fast and accurately by calculating behavior credibility,kinematics credibility,color image credibility and credibility feature weight of joint point.Finally,real-time detection is completed in the virtual scene based on 3ds Max and Unity 3d.Combined with the static gestures and dynamic gesture recognition technology,the design includes the gestures such as starting,pointing to,steering,zooming,waving his hand and stopping.It drives virtual scene to complete real-time changes of corresponding functions and verify the effectiveness of the algorithm.