Research On Real-time Hand Tracking Algorithm Based On Cognitive-behavioral Model Library And Kinect

With the development and popularization of computer and Internet, the human-computerinteraction has become an important part of people’s daily life. In order to realize natural,harmonious and convenient interaction, human-computer interaction technology transfers itsresearch center from traditional computer to human. Gesture is an important means forhuman to express themselves as well as interact with the outside world. As a natural, intuitiveand easy studying interaction means, gesture interaction has a very broad applicationprospects.Gesture interaction based on visual is in line with human natural communication habits.It can realize non-contact and long-range interaction. Thus gesture interaction becomes a hotdirection in the study of natural human computer interaction. Gesture recognition, gesturetracking and gesture interaction based on visual are key technologies to achieve a newgeneration of human-computer interaction. Gesture tracking based on visual gets the videoinformation of hand movement from the input device firstly, then uses the imagesegmentation technology to segment hand for each frame. On the basis of segmentation, thistechnology achieves gesture prediction and tracking, and displays the real-time trackingresults through the3D virtual human model. Based on the analysis and summary on therelated research at home and abroad, this paper does some research on hand segmentationbased on Kinect and hand tracking based on cognitive-behavioral model. The main researchcontents and innovations are as follows:(1) Study the cognitive-behavioral model, to establish the library of microstructure ofhand motionIn particular human-computer interaction environment, the hand motion is regularly.With the guidance of cognitive theory and without human intervention, this paper does theexperiment on the tracking platform based on data glove. During the experiment, the law ofhand motion is analyzed and summarized. To study the cognitive process, interactive stagesand microstructure of hand motion, this paper sets up the cognitive-behavioral model library.Studying the cognitive-behavioral model aims to reveal the interaction patterns and movement characteristics of gesture, and then provides the basis from a new angle for gesturetracking and interaction algorithm design.(2) Based on the microstructure of hand motion, the prediction and sampling based onparticle filter are improved, and a3D hand tracking method based on the microstructure ofhand motion is proposed.At present, hand tracking based on particle filter obtains current3D hand gesture mainlyby two elements, that is, by current frame of gesture images and by status information of theprevious3D hand gesture. Compared with current algorithms, the proposed algorithm of thispaper focuses on behavior analysis and process modeling. Focusing on mathematical linkagesestablishment between microstructure model and the current frame gesture, the status ofcurrent3D hand gesture is obtained not only on the basis of local frame information, butgesture model of overall movement process. After that, a unified and efficient data structureis presented, upon which the particle number can be reduced. This algorithm improves thetracking precision and speed, realizes the real-time and efficient gesture interaction.(3) Hand segmentation based on KinectBecause hand segmentation based on traditional RGB camera has poor robustness, andis affected by light and complex background easily. This paper uses the Microsoft Kinectsensor to obtain depth data, uses the class of OpenNI to get hand location, and then extractsthe field of hand in three-dimensional space. After this, this paper uses the skin colorsegmentation method to process the image. Finally the hand segmentation image can beimproved by denoising, expansion and corrosion process. The experimental results show thatcompared with hand skin color segmentation based on traditional RGB camera, thisalgorithm avoids the influence of light, background and such factors, obtains ideal gesturesegmentation result and improves the robustness of gesture segmentation.