一种空中快速手写连续字符识别方法  

A fast handwritten continuous character recognition method in air

作　　者：刘淑明 巩荣芬[1] 储茂祥[1] 刘历铭 LIU Shuming;GONG Rongfen;CHU Maoxiang;LIU Liming(School of Electronic and Information Engineering,University of Science and Technology Liaoning,Anshan 114051,China)

机构地区：[1]辽宁科技大学电子与信息工程学院,辽宁鞍山114051

出　　处：《辽宁科技大学学报》2023年第6期448-455,共8页Journal of University of Science and Technology Liaoning

基　　金：辽宁省自然科学基金资助项目(2022-MS-353);辽宁省教育厅项目(LJKMZ20220640)。

摘　　要：为改善空中手写连续字符的识别效果,本文提出一种快速的识别方法,利用目标跟踪模型生成空中连续字符的轨迹图像,再利用连续字符识别模型识别轨迹图像。目标跟踪模型采用改进的高帧率孪生网络,连续字符识别模型采用改进的残差循环神经网络。实验结果表明,目标跟踪模型的帧率达到175.5 fps,连续字符识别模型的帧率达到30.8 fps,识别率达到91%。In order to improve the recognition effect of handwritten continuous characters in the air,a fast rec-ognition method is proposed.This method uses a target tracking model to generate the track image of handwrit-ten continuous characters in the air,and then uses the continuous character recognition model to recognize the track image.The object tracking model adopts an improved high frame rate twin network.The continuous char-acter recognition model adopts an improved residual recurrent neural network.The experimental results show that the frame rate of target tracking model reaches 175.5 fps,the frame rate of the continuous character recog-nition model reaches 30.8 fps,and the recognition rate reaches 91%.

关 键 词：空中手写字符 目标跟踪 连续字符识别 人机交互 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]