Interactive Design of 3D Dynamic Gesture Based on SVM-LSTM Model

Introduction

With the development of human-computer interaction technology, a variety of new means of interaction are emerging, so that human-computer interaction is moving towards a more natural, efficient and intelligent direction. Visual hand gesture interaction is one of the main ways of human-computer interaction. Compared with the popular keyboard, mouse and interactive physical touch screen, visual gesture interaction can help users get rid of those constraints, and provide users greater interactive space, more interactive degrees of freedom and more realistic interactive experience. Therefore, the gesture interactive technology has attracted increasing attention around the world, and quickly became a hot research field of human-computer interaction. It has been widely used in virtual / augmented reality, pervasive computing, smart spaces, computer-based interactive games and other fields, such as Leap Motion, Microsoft Kinect and Intel RealSence have quietly stepped into the lives of consumers (Si & Hou, 2016; Wang et al., 2016; Chao et al., 2017).

At the same time, the rapid development of augmented reality is more realistic for the interaction of gestures. At present, augmented reality mainly takes the mobile phone and the head-mounted display as a hardware carrier. The head-mounted display has a stronger immersion than the mobile phone, and the overall experience is even better. Different from the mature touch screen interaction mode on the mobile phone, the head-mounted display will use the natural gesture interaction as the main interaction mode. Because, as compared with conventional virtual reality head-mounted display handle operation, the use of an augmented reality scene is more diverse, more portable equipment requirements. After all, people can accept playing virtual reality games with handle in living room, but do not want to use the handle when operating an augmented reality device outdoors. At the same time, 3D dynamic gestures are one of the most natural gestures of human interaction. Obviously, users want to spend a lower learning costs to use augmented reality device with the natural gestures (Yaoyuneyong, Foster, & Flynn, 2014; Waghmare Amit, Sonawane Kunal, Chavan Puja, & Kadu Nanasaheb, 2014).

Traditional gesture interaction techniques are often based on a statistical learning method or neural network technology, most of these methods have some problems such as poor recognition accuracy, high learning cost and so on. Li et al. (2012) uses Three-point Alignment Algorithm in Kinect to recognize planar gestures with accuracy 84%; Hasan et al. (2014) presents a method for hand gesture recognition using neural networks, which obtain 86.3% accuracy; Dardas et al. (2011) presents a real time system, which includes detecting and tracking bare hand in cluttered background using skin detection and hand postures contours comparison algorithm after face subtraction, and recognizing hand gestures using Principle Components Analysis (PCA) and obtain 90% accuracy; Trigueiros et al. (2013) tries a few classifiers, the highest accuracy classifier is neural networks (accuracy 91%); Lai et al. (2014) presents a method based on convex defect character points of the hand contour, recognizing planar number gestures in real-time with 95.1% accuracy, but planar digital gestures are not natural human-computer interaction gestures; Wang et al. (2013) uses PCA+LDA algorithm in 3d laser scanner to obtain 96% accuracy, but cannot inference in real time.