Research On Vision-Based Hand Gesture Recognition

In recent years,computing technology has improved a lot with each passing day,HCI(Human Computer Interaction)has become an indispensable part of our daily life.As the user experience goes for the leading today,visual-based hand gesture interaction with its obvious advantages like : convenient,natural and friendly,has gradually become an important research branch of human-computer interaction technology.With the emergence of sensors with deep information capture capabilities such as Kinect,vision-based hand gesture recognition research has formed a new direction.As the gesture states are flexible in space and time,the visual-based hand gesture recognition has a high practical value and theoretical significance.In this thesis,based on Kinect sensor in the complex background,we combine static hand state and dynamic finger track as hand gesture characters to carry out research,mainly involve gesture segmentation,fingertip detection,gesture feature extraction,gesture recognition and human-computer interaction gestures design,and finished a series of research and experiment.The main works of this thesis are as follows:(1)Gesture segmentation part,for the face and skin color area in background have bad impact in the hand gesture segmentation,first of all,we use YCbCr elliptical skin color model to split the skin area,and then hand image is corroded to remove the noise.Secondly,the depth information obtained by the Kinect sensor is used to project the skin color region,and the gesture is segmented by the adaptive depth threshold method.(2)Fingertip positioning part,for over coming the existing problems,this thesis base on contour convex theory and introduces a new fingertips positioning method.First,obtains the gesture contour and calculates its approximate polygon.Then,calculates the contour convex vertex to obtain the fingertips candidate points.And finally,through the depth of the neighbor convex defects and the angle between the convex vertex point and neighbor fold points to screen out the fingertips,and access to get the various fingertips position information and fingertips movement characters.(3)Gesture feature extraction part,in order to make the hand gesture interaction more natural and ensure the accuracy of recognition,in the extraction of static gesture features part,we selected the structural characters(fingertips,gestures contour circumference area ratio)and statistics characters(the first four values of the Hu distance)to form the eigenvector of hand gesture.(4)Gesture recognition part,this thesis constructs the multi-valued classifier of support vector machine to recognize the static part of the designed interactive hand gesture and finally recognize the interactive hand gesture and trigger the operation with the finger motion characters.The experiment results show that the hand gestures interaction method in this thesis is natural and flexible.