| With the development and application of computer, the interaction between human and computer becomes more and more intimate. Human-Computer Interaction(HCI) technology arises at the historic moment, and then greatly changed the way people work and live. As an important part of the HCI, gesture interaction also increasingly causes the attention of scientists. How to realize natural, harmonious and real-time gesture interaction has become a hot issue in the research area of gesture interaction.Gesture interaction based on visual does not need people to wear any electronic sensing devices, it only needs one or more cameras used to identify and track hand to complete the gesture interaction. This method meets the natural, harmonious interaction requirements. In the gesture interaction system based on visual, gesture recognition technology decided if the interaction can be completed smoothly or not. Gesture recognition based on vision captures hand images through camera and transmit the image information into the computer. The interaction intention of hand can be judged by computer through image processing algorithm. Then, the computer will execute the corresponding instructions to complete the final HCI. The real-time of gesture recognition also affects the naturalness of interaction. How to realize a real-time retrieval and identification of human hand is the main research goal in this paper. In theory, the consistency between user’s real-time gestures and the corresponding 3D retrieval animation visualization process under the condition of monocular camera is a very challenging work. But it has the important academic value. From the perspective of application background, the 3D interactive interface is an important development trend of future man-machine interface. Therefore, this research has great potential application value on building 3D direct manipulation interface of natural gesture input.This paper is supported by National Natural Science Foundation of China(No.61173079 and No.61472163), and Key Project of Natural Science Foundation of Shandong Province(ZR2011FZ003). We have a research on real-time retrieval method of user basic gestures under monocular camera. In this paper, the main research content and innovation points are as follows:(1) Implement the gesture segmentation method based on skin-color features. Firstly, select YCrCb color space. Then, apply the Gaussian mixture model to simulate the skin color model and complete gesture segmentation by the threshold method. In the initialization process of gestures, we will perform a real-time sampling on gesture skin color area. Then, according to the samples, we can calculate the parameters of Gaussian mixture model. This method increases the adaptability to illumination conditions and reduces the influence of light on gesture segmentation.(2) We propose a feature extraction method based on main direction of gesture. The main direction of gesture is a direction vector from the center of gravity point of gesture to the farthest point of gesture. The purpose of putting forward the main direction of gesture is to guarantee the consistency of gestures feature extraction and solve the problems brought by the gestures in cases such as scaling, translation and rotation. In preprocessing stage, firstly the gesture image after segmentation will be standardized to make the gesture recognition is not affected by gestures scaling. Then, calculate the main direction of gesture on standardized gesture images and build two-dimensional rectangular coordinate system according to the main direction of gesture to extract features of gesture.(3) Propose a concept of gestures binary descriptor. The binary descriptor has great advantage in computation complexity, memory overhead and real-time computation, so the binary descriptor is beneficial to realize real-time computation. To improve the retrieval speed in gesture recognition, the concept of binary descriptor is proposed in this paper. The gesture binary descriptor can be showed in a binary string. Each bit of the binary string is a dimension to represent the number of feature points in the feature area, so the binary string can reflect the characteristic information of the whole gesture area.(4) In this paper, a large number of statistical experiments are performed to prove that dynamic interaction semantic gestures exists certain regularity. In the grasp process, the number of pixel points of hand gestures linearly decrease. In the process of release, the number of pixel points of hand gestures linearly increase. The number of pixel points of hand gestures remains unchangeable in the translation process. The number of pixel points of hand gestures firstly linearly decrease and then linearly increase. Therefore, we can firstly recognize the types of interaction semantic gestures before gestures retrieval, which can reduce the scope of the gestures retrieval and further improve the speed of the gesture retrieval.(5) This paper proposes a gesture retrieval method based on combining gesture binary descriptor and Hausdorff-like distance. Firstly, the main direction of gesture is used to extract features of gesture. Then, interaction types of gesture can be preliminarily recognized by the regularity of interaction semantics gestures. The gesture binary descriptor matching algorithm is used to select some candidate gesture animations from the three dimension gesture animation library. Finally, Hausdorff-like distance is used to retrieve gestures animation which is consistent with user interaction intention from candidate gestures animation. The experimental results show that this method has good real-time performance and also can guarantee higher gesture recognition rate. |
PDF Full Text Request