基于数据手套和神经网络的数字手势识别方法

Digital gesture recognition method based on data glove and neural networks

针对使用数据手套进行数字手势识别时存在个体差异的问题,使用弯曲电阻片设计了数据手套并提出了基于神经网络的数字手势识别方法.首先,在分析测量电路原理的基础上结合弯曲电阻片的特性优选了电路参数,使手指弯曲角度测量的灵敏度最大化.其次,针对用户在手指长度、手势习惯上存在个体差异的情况,提出了一种基于弯曲信号自学习和广义回归神经网络(GRNN)的数字手势识别方法.数据手套信号测试及数字手势试验结果表明,采用优选的电路参数时测量电路的输出振幅最大;在全体评估试验和个体交叉评估试验中,经过自学习预处理后的数字手势识别平均准确率分别为99.2%和96.1%,与未进行自学习处理的识别结果相比分别提高了2.8%和10.7%.在全体评估试验和个体交叉评估试验中,GRNN的识别结果均优于决策树的识别结果.

Aiming at the problem of individual differences in digital gesture recognition based on data gloves,a data glove was designed by using bending resistors and a digital gesture recognition method based on neural networks was proposed.First,a circuit parameter was optimized based on the characteristics of the bending resistor and the principle of the measuring circuit,so as to maximize the sensitivity of the finger bending angle measurement.Secondly,according to the individual differences on the finger length and gesture habits of different users,a gesture recognition method based on signal self-learning and generalized regression neural network(GRNN)was proposed.The results of the data glove testing and the digital gesture recognition experiments show that the output amplitude of the measuring circuit is the largest when the optimized circuit parameter is adopted.In the whole evaluation and the individual cross evaluation experiments,the average accuracy of the digital gesture recognition based on self-learning and GRNN is 99.2%and 96.1%,respectively,which is 2.8%and 10.7%higher than the recognition results without self-learning.The recognition results of GRNN are better than those of the decision tree in the whole evaluation and the individual cross evaluation experiments.