Visual Gesture Recognition Systembased On Android

Main operation process of current mobile devise is touch control. However, under special application scenes, such as the oil field or the mining area and so on, due to the high density of dust, oil and noise, it is inconvenient to carry out touch control and voice control. Gesture, as a natural and direct man-machine interaction method, is more appropriate for these areas. Therefore, it is very significant to develop the gesture identification in the mobile devise. As early as 1993, Thomas had proposed to carry out the man-machine interaction of gesture with the method of glove. Although the identification was accurate, the existence of glove setting was still not friendly to users. Latter, the visual way was proposed to identify the connotation of gesture, including the template matching method and mathematics statistical method. The development of gesture identification is becoming increasingly sound. However, the majority of the above mentioned methods are the gesture identification based on the phenomenon that the image is at the PC end. There is still blank in the identification based on the mobile devise end. There are big differences between mobile devises and traditional computer in terms of the performance and it is possible that methods being applicable in the traditional computer are not feasible for the mobile devise. To guarantee the high identification accuracy, it must design more precise identification process and thus improve the algorithm complexity. However, since the performance of mobile devise is low, the design of high accuracy would seriously affect the timeliness of identification. In the contrary, if more attention is placed on the timeliness, it would be difficult to guarantee the accuracy of algorithm. How to achieve a balance is a difficult point.This study, drawing on prior visual gesture recognition technologies, particularly focuses on the geometrical characteristics of gestures and there is proposed a visual gesture recognition method being able to run smoothly on mobile devices. The original gesture images are captured using a mobile device camera, the binarized gesture images being obtained by means of skin color recognition. Geometrical characterization analysis of binarized gesture images helps to determine useful geometrical characteristic values so that the gesture meanings are found by characteristics matching. The gesture information is then interpreted as instructions that control and operate the device. The test platform used by this paper was Android and the testing made clear that with an Android device this method has an operation cycle of about 120 ms and a correct rate close to 90%, guaranteeing the real-timeliness and accuracy of gesture recognition.