Research And Application Of The Gesture Modeling Algorithm

With the rapid development of computer hardware and software technology, the gesture recognition which fits the interpersonal communication habits is made possible. It has become a challenging interdisciplinary research topic in that the vision-based gesture recognition technology is related to computer vision, image processing, pattern recognition, and gestures themselves are diverse, ambiguous and different in time and space. This paper mainly touches on the modeling process of gesture recognition, including the analysis of the specific gestures and the gesture recognition.The final result of gesture recognition algorithm based on the traditional Hidden Markov Model (HMM) is decided by the one with the largest output probability model. However, due to the little movement range in gesture trajectory of the senders and the similarity of their gestures, the output probability of different HMM models may be very close to each other. There is the possibility of false identification based on the maximum probability alone. Besides, gesture recognition based on Support Vector Machines (referred to as SVM) which use the overall trajectory numerical features to identify, ignores the various changes in the middle of gesture trajectories. The HMM model, which uses the contextual pattern recognition, expresses the category similarity greatly. Meanwhile, the SVM model, which excels in classification, reflects the class differences. Therefore, the integration of the two just makes up the deficiencies of each other.Since the HMM model has good time series modeling capabilities and the SVM model has the excellent performance in terms of classifying the finite sample, this paper presents a fusion model—HMM_SVM gesture recognition model based on trajectory segmentation. The basic idea is to divide the gesture trajectory, which is obtained from the gesture feature extraction process, into two parts. Then they are put into the trained 16 HMM model respectively to calculate the 16 maximum likelihoods. If the outputs meet certain conditions, the outputs should be re-classified using SVM. This can make the gesture recognition more accurate and improve the accuracy of gesture recognition to a certain extent. This research also realizes a prototype real-time gesture recognition system that includes gesture image capture, image pre gesture, gesture segmentation, gesture feature extraction, gesture trajectory recognition process. In the gesture image pre-processing stage, this paper provides a local adaptive image enhancement algorithm based on the ILAE algorithm, which can lay a good foundation for subsequent image segmentation and feature extraction of hand gestures. The gesture recognition experiment results show that the average recognition rate of 8 representative gesture tracks based on HMM model alone is 93.25%, while the average recognition rate of those representative gesture tracks based on the HMM_SVM integration gesture recognition model is 95%. The latter has improved the efficiency of gesture recognition to a certain extent.Finally, the paper points out some of inadequacies in the current works and offers prospects for the follow-ups.