A Study Of Hand Gesture Identification Based On Visual Neural Network

With the rapid development of virtual reality technology, a new people-centered way of new human computer interaction becomes extremely active to achieve a more natural demand. In this way, interaction is realized by means of identifying people’s eyes、 fingerprints、face expression、head detection posture and gesture and so on, which is more simple and convenient than the interactive way using traditional mouse keyboard, as a result, it becomes the research hotspot in recent years in the world. Among them, gesture recognition has more applications in man-machine interaction technology, as a special way of recognizing biological characteristics, and this kind of interactive way is in line with human communication habits. The study of gesture recognition based on vision is the trend to multi-disciplinary research topic with chanllenges because human hands are complicated deformation, visual obstacles, ambiguity and diversity, and so on.The based on analyzing development situation and application prospect in gesture recognition technology, an implementation scheme of gesture identification system is proposed using algorithms of visual neural networks. First of all, in gesture image preprocessing part, original sequence gesture images from camera is input, and moving gesture is captured by using a kind of axonal delay mechanism based on the spiking neural network model, combined with the human body skin color information to segment the hand gesture from complex background, so as to built sample gesture library; Then, in the process of gesture’s feature extraction, considered the uncertain of scale and rotations, a set of feature extraction algorithms are proposed, including geometric characteristics and distance direction histogram feature extraction algorithm based on spatial distribution, and then the multiple characteristics extracted are used to feature fusion; Next, a characteristic vector which combined the features is used to train the classifier after constructing one-against-one SVM multiclass classifier model, to complete the ten predefined static digital gestures recognition; Finally, in the part of gesture tracking, Camshift method is utilized to obtain the trajectory of the hand and achieve betimes tracking. Based on these, we develop multiple applications such as controlling the mouse, drawing and so on, to realize the dynamic application interaction between user gestures and computer.Experiment results show that the feature extraction algorithm proposed in this paper can characterize the expression features of gesture images effectively. The feature extraction algorithm is combined with the feature fusion and SVM multiclass classifier organically to improve the gesture recognition accurate rate.