采用RBF神经网络求解反向条纹的研究  

Inverse Fringe Solved with RBF Neural Network

作　　者：陈云富[1] 李勇[1] 张海花[1] 蔡元元[2] 

机构地区：[1]浙江师范大学信息光学研究所,浙江金华321004 [2]中科院成都信息技术有限公司,四川成都610041

出　　处：《光学与光电技术》2010年第5期37-40,共4页Optics & Optoelectronic Technology

基　　金：国家自然科学基金(60702078)资助项目

摘　　要：提出了一种利用RBF神经网络来确定摄像机和投影器坐标映射关系的方法。首先在投影器坐标系中将数据分为若干个16×16的子区域,然后以(l,m,lm,l^2,m^2)为输入层的5个神经元(其中l、m为投影器像素坐标),以摄像机像素坐标i为输出层的神经元,建立RBF神经网络。利用RBF神经网络求解在投影器坐标系中摄像机像素坐标的分布模型,最后得到投影器像素点对应的摄像机像素坐标值。计算机模拟和实验结果表明,与已有的算法相比,该方法能更有效地提高反向条纹投影的求解精度。为反向条纹的求解提供了新方法。A method based on RBF neural network is proposed to map the camera coordinates to proiector coordinates system. Firstly, the data are separated to several regions with size of 16 × 16 points in projector coordinates system. Then, the RBF neural network is constructed by regarding （l, m, lm, l2 , m2） as nerve cells in input layer and camera coordinates i as nerve cells in output layer. The distributing model of camera coordinates in projector coordinates system is obtained with RBF neural network. Finally, the camera coordinates corresponding to projector coordinates are gained. Computer simulation and experiments reveal that the accuracy of inverse fringe is increased with the proposed method.

关 键 词：径向基函数神经网络 反向条纹 二次三项式插值 工业质量控制 

分 类 号：TN247[电子电信—物理电子学]