Research On Interaction Between Pose Calculation And Static Gesture Recognition In Mobile Enhancement Reality

With the popularization of mobile devices, such as convenient carrying and low energy consumption of smart-phone, it naturally becomes the best platform for mobile augmented reality (MAR) application, and MAR based on smart-phone has important theoretical significance and practical value.For traditional augmented reality systems, there are defects as follows:high time consumption in corner detection, low pose tracking precision and non-real-time. Traditional augmented reality systems are not suitable for direct transplanting to mobile terminals. In order to solve the above shortcomings, considering limited computational resource of smart-phone, a fast four-corner detecting algorithm, which based on image segmentation’s flood-fill process and bounding box’s geometric features, is proposed to reduce computation time. Combined with marker-board, even when the marker is partial occluded, it could also detect the marker. In order to quickly and accurately compute camera pose, the homography matrix is firstly optimized via Levenberg-Marquardt iterative algorithm, then the jitter caused by inaccurate camera pose is solved. The experiments show that fast corner detecting algorithm and optimal method of homography matrix are faster and more accuracy than that of other methods used in ARToolKit and ARTag. Therefore it is feasible for smartphones with low computational resource.On the basis of above fast corner detecting algorithm, a new interactive technology of augmented reality is studied for static hand gesture recognition. Firstly, hand segmentation is done via skin color detection, morphology processing and contour extraction. Secondly, the static hand gesture recognition algorithm, which is based on fingertip-detection using curvature information of hand-contour, is introduced and implemented. Finally, the hand-contour can be represented as feature vectors using Fourier descriptors, then the cluster centers of hand-contour are calculated by using K-means clustering, the recognition result is determined by the distance between a hand-contour feature vectors and the cluster centers. Reliable recognition rate is verified by the experiments.