Theoretical Summary And Gesture Recognition Study Of Conditional Random Fields

In recent years, human-computer interaction based on natural human form is a very active area of research. Class is the highest stage of human interaction with computers in the pursuit of human interaction, not only allows users to computer performance in the face of more natural, but also to allow the intelligence of the computer to a new level. Currently, this form of human-computer interaction studies have focused on the gestures, posture, language, facial expressions and so on. Computer vision image processing means for extracting feature information of the person and the data will be described, and then the target is identified by pattern recognition methods. Among them, the most important stage is the target recognition, due to the characteristics of the information people have high dimension, characterization of complex, difficult to represent completely and so, so many methods using a model of intelligent computing. Artificial selection and build an appropriate model, trained by feature information optimization parameters of the model, and then through the model to determine the target. This paper CRFs model starting from its structure, the probability that the parameter estimation, marginal inference model described in detail issues such as evolution, and gesture recognition monocular camera, for example, describe the use of conditional random model approach. Specific contents are as follows:First, the CRFs model itself, from the class point of view it is an undirected probabilistic graphical models, from the point of view that it is a probabilistic description of the discriminant model. The model itself is dependent on the decision of its use of space and features, in2001on the basis of maximum entropy model Laflerty and hidden Markov model, we propose a conditional random modelThrough a simple first-order conditional random structure, constructed in the form of the potential function, description and parameters of the marginal probability estimation method and successfully applied it to the object recognition problem. Subsequently, the conditional random field model of the structure to become a flexible task, many researchers extended its application to identify the more complex problem, which is studied in this paper will lay the foundation.Then, for the single camera gesture recognition, we propose a recognition method based on implicit conditions with the airport.The method is based on decomposition of the hand movement vector locus consecutive frames, and each frame is described by the vector direction of the standard eight-direction matching, the length of the interframe vector dimension as the characteristic direction, the data processing method, the feature Statistics conducted on the dimensions and normalization, then the probability of an implicit form of identification with the structural conditions of the airport verify assumptions. As an application of the experimental model used CRFs example, the model for subsequent research and identification of problems and lay the foundation also verified CRFs model performance and features.Experimental results show that the proposed conditions with gesture-based implicit recognition method airport is effective, has a high recognition accuracy. Model for the in-depth study to identify the problem and lay a good foundation.