查询结果:   王红全,淮永建.基于Leap Motion手势识别方法在树木交互的应用[J].计算机应用与软件,2018,35(10):153 - 158.
中文标题
基于Leap Motion手势识别方法在树木交互的应用
发表栏目
人工智能与识别
摘要点击数
670
英文标题
APPLICATION OF GESTURE RECOGNITION BASED ON LEAP MOTION IN THE FIELD OF TREE INTERACTION
作 者
王红全 淮永建 Wang Hongquan Huai Yongjian
作者单位
北京林业大学信息学院 北京 100083     
英文单位
School of Information Science and Technology, Beijing Forestry University, Beijing 100083, China     
关键词
Leap Motion 手势识别 特征提取 树木交互 LSTM
Keywords
Leap Motion Gesture recognition Feature extraction Tree interaction LSTM
基金项目
国家自然科学基金项目(31770589)
作者资料
王红全,硕士生,主研领域:计算机应用技术。淮永建,教授。 。
文章摘要
随着人们对更真实体验和更智能交互的追求,虚拟现实技术与深度学习技术成为当前的研究热点。利用Leap Motion设备,结合机器学习算法进行手势训练与识别并应用到树木交互场景中。利用Leap Motion设备采集静态手势和动态手势数据并对其进行特征分析。针对静态手势提出使用SVM算法进行分类,通过对采集5种静态手势的1 000个样本训练与识别,平均识别正确率达到96.3%;针对动态手势提出使用LSTM模型对手势时间序列进行处理,通过对5种动态手势进行评估后发现其平均准确率达到高达92.6%。将训练的手势识别结果应用到树木交互场景中,分别选取3种静态手势和动态手势,实现对场景中的树木模型的交互。实验结果表明,使用LSTM模型手势识别方法可以实现更高的动态识别准确率;结合静态和动态手势识别算法,通过匹配离线模型训练的手势库,可以提高用户对树木的交互准确性。
Abstract
In pursuit of more real experience and intelligent interaction, virtual reality technology and deep learning technology have become the current research hotspots. In this paper, we used Leap Motion device and machine learning algorithm to train and recognize gestures, and applied them to tree interaction scene. We utilized Leap Motion device to collect static and dynamic gesture data and analyzed the characteristics of them. SVM algorithm was used to classify the static gestures, and the average recognition accuracy was 96.3% by training and recognizing 1 000 samples of 5 static gestures. For dynamic gestures, LSTM model was utilized to process the time sequence of the gestures, and the average accuracy rate was up to 92.6% by evaluating the 5 dynamic gestures. We applied the results of trained gesture recognition into the tree interaction scene. 3 static gestures and dynamic gestures were chosen to realize the interaction of the tree model in the scene. The experimental results show that gesture recognition method based on LSTM model can achieve higher dynamic recognition accuracy. This paper combines the static and dynamic gesture recognition algorithm to improve the accuracy of user interaction with trees by matching the gesture library trained by the offline model.
下载PDF全文