基于时频域指纹映射的中国书法运笔姿态与动作的无线感知与识别方法研究

There is a strong need for a new method to guide calligraphy learning so that the policy “Calligraphy is into the classroom” can be carried out. However, traditional demonstration teaching approach face the limited time of course, a number of students, shortage of teachers and etc. To cope with these challenges, this project tries to sense and recognize the motion features of calligraphy brushstrokes by using wireless signals. We try to use combine RSSI information、amplitude of CSI information and phase of CSI information to improve recognition accuracy. To enhance the robustness, transfer learning is used to deal with the handwriting diversity problem for different writers. LOF algorithm is used to disassemble the signals of motion features and time-frequency domain analysis is used to extract the features from the disassembled brushstroke signals so that the computational complexity and storage space can be reduced. HMM and fuzzy cluster algorithm are utilized to build an objective evaluation model to judge and correct the hand-writing copy, the efficiency of hand-writing copy is improved accordingly. We are planning to verify the practicability and efficiency of the solutions in real scenes by doing experiments in Forest of Steles Museum and carrying out related experiments. The scientific essence of this project introduces transfer learning and probability theory in behavior recognition, seeks the intrinsic relationship between wireless signals and motion features of calligraphy brushstrokes. The research findings of this project will provide theoretical for a new generation of hand-writing copying and will have a positive effort on it.

“书法进课堂”政策的落地，迫切需要新的有效的书法学习与指导方法。本项目试图利用无线信号对书法运笔姿态与动作进行感知与识别，从而解决传统观摩式教学面临的课时短、学生多、师资少等挑战。结合RSSI及CSI的幅值相位信息提高书法运笔姿态信号的识别准确率，引入迁移学习解决不同书写者的书写多样性问题，增强鲁棒性；利用局部异常因子检测算法对书法运笔姿态和动作信号进行拆解，结合时频域分析方法对拆解后的信号进行特征提取，以降低识别的计算开销，避免因样本库过大而造成存储空间消耗；利用隐式马尔可夫模型和模糊聚类算法建立临摹效果评测体系，对临摹效果进行评判及纠正，提高临摹效率。项目将通过碑林博物馆以及相关实验环境验证所提方法在真实场景下的有效性。其科学实质是在无线信号行为识别中引入迁移学习、概率统计理论，寻求无线信号与书法运笔姿态之间的内在规律，研究成果为新一代书法临摹应用提供有价值的理论支撑和应用参考。

书法是中国乃至世界范围内一种流行的、备受推崇的艺术形式，是中国文化的重要载体。随着“书法进课堂”政策的落地，急需新的高效的书法学习与指导方法。本项目针对传统观摩式书法教学面临的课时短、学生多、师资少等问题，研究了通过先进感知技术识别书法运笔姿态和动作进而高效准确的指导初学者练习书法等科学问题。项目提出的基于迁移学习的跨场景识别和基于集成学习的大规模无线行为识别方法，有效地解决了WiFi感知中不同场景采集训练数据的开销问题和大规模场景中系统性能迅速下降的问题，为解决书法运笔姿态识别中的书写多样性、空间多样性问题提供了理论参考；项目提出的基于目标追踪算法的图形密码锁攻击方法，能在5次尝试中成功破解90%以上的智能手机上的图形密码锁，该研究为连续书法运笔姿态与动作信号的拆解方法提供了理论参考；本项目针对中国书法运笔姿态与动作的感知，分别提出了基于计算机视觉、基于WiFi信号、基于RF技术的书写姿态评估方法，能准确的跟踪笔刷的运动轨迹，将其与知名书法家的示例模板进行比较来指导书法初学者的运笔姿势，很好的解决了传统现场观摩式书法教学面临的专业师资队伍严重紧缺的问题。该项目的多个研究成果先后发表在ACM Mobicom 2018，ACM CCS 2018, IJCAI 2018，IEEE TON等高水平国际会议和期刊上，并被美国斯坦福大学、德国萨尔大学、清华大学等多个国际国内知名高校引用。相关成果被新华社、华商报、福布斯、泰晤士报等国际国内主流媒体广泛报道。本项目的研究对“书法进课堂”政策的实施和发扬中国传统文化有着重要意义。

内容获取失败，请点击重试