Research And Implementation Of Gesture Recognition System In Natural Scenes

As an important part of human-computer interaction,gesture recognition affects the naturalness and flexibility of human-computer interaction.At present most researchers have focused on gestures recognition,often to simplify the gestures background,and in the context of a single use of the study algorithms will gesture segmentation,and then takes the recognition method of commonly used gestures to express the meaning of out through the analysis of the system but in the real applications,the gesture is usually in a complex environment,for example: the light is too light or too dark there are more gestures gestures from the acquisition device from complex background factors such as different.These problems has not been resolved,and can also be difficult to solve in the future so need the researchers are now expected to solved problems in certain circumstances,and then through the combination of a variety of methods to achieve the suitable for different gesture recognition under complex environment,the humanized human-computer interaction for gesture recognition research and in the future.In this paper,we have two aspects in total:1.First of all,we solved the problem of low quality pictures obtained by ordinary cameras.Due to the influence of light intensity and camera pixels,the pictures obtained by our camera will not always get high-quality and clear pictures,which will increase the difficulty in our gesture recognition process.In order to solve this problem,we propose a multi-scale feature fusion attention network to solve the problem of low quality and fuzzy image acquisition.The proposed structure not only greatly improves the accuracy of our gesture recognition,but also guarantees the real-time performance of our gesture recognition.2.For gesture recognition,we first need to conduct gesture analysis,which is one of the key technologies to complete the gesture recognition system.Through gesture analysis,the shape feature or motion trace of gesture can be obtained.The shape and trajectory of gesture are important features in dynamic gesture recognition and are directly related to the meaning of gesture.The main methods of gesture analysis are as follows: edge contour extraction method,centroid finger multi-feature combination method and knuckle tracking method.Edge contour extraction is one of the commonly used methods in gesture analysis.Secondly,gesture recognition is the process of classifying the trajectory(or point)in the parameter space of the model into a subset of the space,which includes static gesture recognition and dynamic gesture recognition.From the point of view of the realization of gesture recognition technology,the common gesture recognition methods mainly include: template matching neural network method.