Research On Complex Gesture Recognition Technology Based On Kinect

With the continuous development of human-computer interaction technology,the novel interaction methods emerge in an endless stream.The gesture recognition technology with its low cost,good flexibility and strong practicability,has become a hot research in recent years.The gesture recognition based on visual technology is affected by illumination and noise,which limits its application.The Kinect sensor can obtain the depth information of the space during acquiring 2D images at the same time,which brings a new direction for the research of gesture recognition.This paper uses the Kinect2.0 sensor,to recognize the dynamic hand gestures with handshape change,which includes hand image segmentation,hand features and gesture feature extraction,gesture classification and recognition.First of all,for the segmentation of hand image,the Kinect skeleton tracking technology and the depth information are combined to eliminate the influence of background and illumination on the image segmentation.The obtained binary image of hand is processed by morphology and then the outline of hand image is extracted by edge tracking algorithm.Then the extraction of gesture features,which includes static hand features and trajectory features.The static handshape is extracted by Hu feature method and then encoded using K-means clustering algorithm.The trajectory is extracted by direction angle feature,and then the feature is quantized to spherical 14 direction,hand feature encoding and direction feature encoding are obtained respectively to construct the handshape feature sequence and direction feature sequence.Finally,the classification and recognition of hand gesture,a HMM-NBC model,combined with Hidden Markov Model and Naive Bayesian Classifier,is proposed for gesture training and recognition.For custom 10 dynamic gestures,the average recognition rate reached 88.4%.