| Human-computer interaction is an important part of computer science research.In recent years,as the functions of computers have become more and more powerful,the operation methods of human-computer interaction have also become more and more convenient,and people have increasingly favored non-touch human-computer interaction methods.Among them,gestures are flexible and changeable.They are easily controlled by people.They are widely used in human-computer interactions.For example,handicapped persons with disabilities and empty-nests,children with autism,and deaf-mute persons can be provided with gesture recognition.Safeguards and convenience.This article uses the Kinect camera to capture image information and perform static gesture recognition and dynamic gesture recognition,respectively.Static gesture recognition mainly consists of three parts: hand gesture segmentation,feature extraction and classification recognition.Traditional skin-based hand gesture segmentation will be affected by light and skin-like skin background.Segmentation based on depth threshold will be interfered by same-depth objects.Therefore,this paper combines the two segmentation methods to compensate for similarities in skin color background and depth.Disturbance from the object effectively gets the gesture area.Then extract the fingertip number and Hu moment feature from the segmented gesture area,and extract the number of fingertips to detect the effective farthest point using convex hull defect.Finally,the feature vector of fingertip number and Hu moment is taken as the input value,and the SVM classifier is used to classify and identify static gestures.Dynamic gestures include both hand and trajectory features.A dynamic gesture cannot be fully described only by the hand or trajectory,and it is easy to recognize similar errors.Therefore,this paper combines hand-type recognition and trajectory recognition,that is,a more detailed description of dynamic gestures.First,the gesture region is segmented by the RGB-D method.For the hand shape change feature,a keyframe extraction based on the gesture area change is used for the segmented gesture portion,and a standard static gesture image capable of describing the dynamic gesture shape change isobtained using SVM.The static gesture recognition is performed,and the recognition results are combined and arranged according to the key frame extraction order.For the motion trajectory feature,centroid extraction is performed on the segmented gestures,and the gesture centroid of each frame is connected to form a motion trajectory,and the detection trajectory preprocessing of the gesture is completed,and the motion trajectory is classified and identified by the HMM algorithm and the trajectory direction angle.The final dynamic gesture recognition result is a combination of the hand shape and the track recognition result.Finally,the static and dynamic gesture recognition results are programmed to control the NAO robot motion.Experimental results show that the gesture recognition algorithm in this paper can effectively identify static and dynamic gestures and achieve human-computer interaction. |
