基于BP神经网络的鼠标轨迹识别技术  

Mouse Trajectory Recognition Technology Based on BP Neural Network

作　　者：陈喆[1] 周雷[1] CHEN Zhe, ZHOU Lei （Chengdu University of Technology, Chengdu 610059,China）

机构地区：[1]成都理工大学,四川成都610059

出　　处：《电脑知识与技术》2013年第1期130-132,共3页Computer Knowledge and Technology

摘　　要：鼠标与电脑交互通常采用点击的方式,而由鼠标画出特定的轨迹,由轨迹识别算法识别出所属类别,并执行相应的命令,即鼠标手势,这是一种更高效的交互方式。轨迹识别属于分类问题,该文采用BP神经网络技术实现鼠标轨迹的识别。针对人工生成训练样本耗时、低效的缺点,该文提出了使用原型轨迹的变换版本扩充训练样本集（提取角度特征,给原型轨迹加噪声）。最后训练11种轨迹,并对BP神经网络识别性能做测试,实验结果表明识别率达到92.9%。Trajectory recognition is a classification problem, in this work, the BP neural network technology is adopted to achieve the recognition of the trajectory of the mouse. For artificially generated training samples is time-consuming and ineffi- cient, this paper proposed a novel method to expand the training set using transformed version of the original input patterns （ex- tract angle feature, and then add noise to the original trajectory）. Finally, training 11 types of trajectories, and test the performance of the BP neural network, the experimental results show that the recognition rate of 92.9%.

关 键 词：BP神经网络 鼠标轨迹识别 训练样本生成 

分 类 号：TP18[自动化与计算机技术—控制科学与工程;自动化与计算机技术—控制理论与控制工程]