Resarch On 3D Hand Tracking Based On Cognitive Model

Research on three-dimensional posture estimating and three-dimensional hand tracking in video sequences is the fundamental and key project which covers various cross-disciplines,such as computer vision, Human Computer Interaction(HCI), estimation methodology.So,it is most significant to conduct thorough research on three dimension(3D) hand tracking for boosting theories and applications of intelligent HCI and promoting the application and researches of relative disciplines.Under the circumstance of single-view, the main purpose of appearance-based tracking and model—based tracking is to obtain the 3D hand gestures and their positions corresponding to the frame images, and the main method of tracking is getting the 3D hand gestures at time k by prediction and filtering according to the 3D hand gestures at time k-1 and the frame images at time k. And the main feature of this method is by establishing the mapping relation of features of 3D hand gestures and the features of frame images, and getting the 3D hand gestures whose model error is smallest. It is a problem of searching and matching in high dimension space from its very essence.Because the hand gesture, on itself, is complex, and is characterized by multiformity in appearance, ambiguousness in semanticness and differences in space-time, vision-based tracking and recognition to the human hand is a challenging project. Besides, the hand gesture is characteriatic of high degree of free(DOF), assumed that each DOF only has two change trend, and the computational complexity can get to 2 33 for the hand gesture tracking while the hand gesture including 33 DOFS.And the computational complexity is too high to reach the goal of real-time tracking.To address these problems, a novel hand tracking based on cognitive model is proposed which combine computer science and cognitive psychology relevant discipline.The paper researches on the folling key problems in 3D hand trcking:(1) Gesture recognition in complex background based on distribution features of hand.Gesture recognition is one of the key technologies in advanced human-computer interaction. In this paper, a hand-distribution-features-based approach to hand gesture recognition is presented. In this method, search-window is used to selecting the valid hand gesture, and extracting the hand distribution features which including the destiny distribution feature and the figures features; finally we integrated all the features to calculate similarity distance. In this paper random sampling is introduced which can improve recognition rate, while the hand and the face all in our Video Sequence Images, we can recognize the hand gestures still because the using of search window. The experimental results show that if the ambient light is relatively stable, the algorithm proposed in this paper is real-time with strong robustness, and is invariant to rotation, scale and translation.(2) A new method of hand tracking which based on cognitive model is put forward.In natural Human Computer Interaction, there are the psychological activities and psychological characteristics of operators everywhere, and the mental activity directly affects the human hand moving. And in this paper, we cognitive psychology, behavioral sciences as its rationale to get the features of hand moving and the operators'psychological characteristics. First, we use the method of observation, experiment, psychology, oral reports and other analysis methods to study characteristics of the operator's cognitive and psychological characteristics of gesture movement which occurred in a specific human-computer interaction, and trained to different operators with a complete implementation movement, by the help of a virtual assistant platform for getting the motion data, and then understanding the data curve fitting, to make concrete gestures movement characteristics, digital, laying the groundwork for the follow-up sample. And on this basis, we analyze the variation tendency of the data, classify the data of the gesture by way of probability, for the particle filter sampling method provides an efficient, unified model of mathematical knowledge, as a basis for the sampling method can avoid blindly searching in high dimensional space, and reduced the dependence on the dynamic model while tracking the hand gesture, but more trust the priori knowledge provide by the cognitive model. The experimental results show that compared with the traditional filtering algorithm, our hand tracking algorithm which based on cognitive model can use less running time to achieve higher precision.