面向大类别的空中手写中英文识别技术研究

项目名称： 面向大类别的空中手写中英文识别技术研究

项目编号： No.61772495

项目类型： 面上项目

立项/批准年度： 2018

项目学科： 其他

项目作者： 王伟强

作者单位： 中国科学院大学

项目金额： 17万元

中文摘要： 空中手写作为一种新型的基于视觉的人机交互方式，开展相关关键技术的研究（准确识别所书写的字符或符号）具有重要的研究意义与实用价值。本项目将深入开展面向大类别数的空中手写字符（汉字，英文，数字以及特定功能符等）识别技术研究。本项目具有如下特色：（1）我们率先使用Leap Motion三维传感器来捕捉跟踪手指的运动轨迹；（2）除了数字，英文字母外，我们将针对更复杂的大类别数(约6900类)的汉字字符识别开展研究；（3）将率先开展具有挑战性的中英文句子级的连续空中手写输入的研究；（4）支持的识别内容除覆盖中英文、数字外，还将开展一些代表特定命令功能的手势识别技术的研究, 以配合中英文的高效输入以及面向特定应用的人机交互。本项目将探索基于手指运动轨迹与手的姿态相结合的新型人机交互技术，拟通过发表高水平科技论文吸引国际学者对该问题的关注，并通过建造公开大规模高质量的相关数据集推动国际上相关技术的研究。

中文关键词： 空中手写字符识别；深度学习；人机交互

英文摘要： The in-air writing, as a vision based human-computer interaction way, has a significant value in theory and practice. The project aims to explore the methods to recognize in-air handwritten text with a large number of pattern classes, including Chinese text, English text, digits and signs with special functions. Compared with previous works, the project has such characteristics: (1) We first exploit the Leap motion 3D-sensors to capture and track the motion of fingers; (2) Besides English character and digits, we first start up the research on more complex Chinese character recognition （with about 6900 pattern classes） ; (3) We first investigate the techniques to implement the continuous input of Chinese and English sentences in the air ; (4) Besides recognizing Chinese, English, digits etc., the recognition of hand gestures which represent special functions, is also investigated to make the input of Chinese and English more efficient or support the human-computer interactions in a specific application. This project will explore the novel human-computer interaction technique based on the motion trajectory of fingers and hand gestures. We expect that our research work should draw the attention of other researchers by reading our high-quality scientific paper, and we also hope the big dataset constructed with high quality by us can push the related world-wide research to a high level.

英文关键词： in-air handwritten character recognition;deep learning;human-computer interaction