Research On Algorithm For Gesture Recognition Of Hand And Face Interactive Behavior

The recognition of hand-face interaction is an important and challenging research subject during sign language recognition and abnormal behavior recognition. There have a vast amount of hand-face interaction vocabulary and syntactic of sign language to recognition, As for the identification of abnormal behavior such as smoking, gestures are often used as a criterion to supply cigarette smoking detection during complex background.The research of hand-face interaction mainly facing with hand close to face, similar tone and texture of skin, as well as the limited number and quality of the experimental dataset.Therefore, the research of hand-face interaction gesture recognition based on machine vision has a very important and far-reaching significance.In this paper, we utilize the existing hardware to capture and establish RGBD dataset of hand-face interaction, and a new gesture detection method of Kernel Fuzzy C-Means(KFCM) cooperating with Bias Field Correction Based Multiphase Level Set(BFMLS)model is proposed based on the fusion of color and depth information. Firstly, using KFCM algorithm to get rough segments of hand adjoin to face occlusion images, and further enhance the images. Then, using the enhanced result to initialize the level set function to solve the missing pixels in the hand region.In order to improve the accuracy and reliability of the feature vectors, the histogram of oriented gradient features is improved based on depth information. For the depth information has the characteristics of complete and accurate description of the spatial information of the object, and the histogram of gradient has the characteristics of accurate description of the direction of gradient of the local area in the image. Therefore, the histogram of oriented gradient of depth can efficiently describe the shape and internal texture of gesture, and it is less affected by the environment, has strong reliability and adaptability.Hand-face interaction recognition is still at the primary stage, taking into account the less prior knowledge. This paper mainly using the support vector machine as the classifier of Hand-face interaction recognition, this is because of support vector machine has thefollowing advantages: strong generalization ability, good stability, and excellent classification rate. comparing with different gesture segment methods and feature extraction methods with self built database, the result shows that the proposed gesture segmentation algorithm and the improved histogram of gradient feature have the effectiveness and superiority in dealing with the recognition of hand over face occlusion in the near mode.