仿人手感知的纹理识别与触觉再现方法研究

Research on texture recognition and tactile representationmethod based on humanoid tactile sensing

提出了一种仿人手感知的纹理识别及触觉再现方法。采用新型纹理检测装置以仿人手触觉感知方式采集纹理,对采集到的纹理信息进行主成分分析(PCA)降维提取特征,搭建多层感知(MLP)神经网络对6种纹理信号进行分类,识别率高达98.8%。在此基础上,基于采集数据的统计特性,采用正态分布振动模型再现纹理触感,并通过滑动窗口动态修正力学模型,强化局部纹理特征。通过对比实验,所提再现方法在对布料等细腻纹理的再现效果上优于传统基于图像信息的再现方法。

A method of texture recognition and tactile representation using a new texture detection device based on the process of finger tips sensing object is presented.The collected texture information is extracted by principal component analysis(PCA)dimension reduction,and a multi-layer perception(MLP)neural network is built to classify 6 texture signals,with the recognition rate as high as 98.8%.On this basis,the normal distribution vibration model is used to represent the texture tactility based on the statistical properties of the collected data,and using sliding window to modify mechanics model dynamically in order to strengthen the local texture feature.Through the comparison experiment,the algorithm in this paper is better than the traditional algorithm based on image information in the representation of fine texture such as cloth.