基于小波神经网络的复杂三维物体测量  

Three-dimensional measurement of complex object based on wavelet neural network

作　　者：蔡义祥[1] 陈文静[1] 赵玥[1] 许罗鹏[1] 

机构地区：[1]四川大学光电子科学技术系,成都610064

出　　处：《四川大学学报（自然科学版）》2010年第4期763-767,共5页Journal of Sichuan University(Natural Science Edition)

基　　金：国家自然科学基金(60677028)

摘　　要：将小波神经网络引入基于结构光投影的复杂物体三维面形测量.在测量过程中,利用小波函数的时频特性及变焦特性和神经网络强大的函数逼近功能,得到离散条纹图的连续逼近函数,从中解出物体的相位信息,获得物体的三维面形分布.应用小波神经网络,在结构光投影条件下,只需要获取一幅条纹图,便可以完成复杂物体的三维面形测量.该方法相比传统的傅里叶变换轮廓术方法,不存在滤波操作,具有更高的灵敏度,在条纹图存在阴影的情况下,能更准确获得物体的相位信息,更加适用于恢复复杂物体的三维面形.模拟及实验均验证了该方法的可行性.The wavelet neural network has been introduced into the reconstruction of the complex threedimensional（3D） object based on structured light projection. In the method, the wavelet with time-frequency characteristics and zoom features and the neural network with powerful function of approximation is used to get the continuous approximate function and draw phase distribution of the object. As a result, the wavelet neural network method based on structured light projection needs only one deformed fringe pattern to reconstruct the tested object. Compared with the Fourier transform profilometry, the wavelet neural network without filtering process and with high sensitivity can demodulate more useful phase from the fringe pattern with shadow. Therefore, this method performs better than Fourier transform profilometry in the three-dimensional shape measurement of complex objects. The feasibility of this method is validated by computer simulations and experiment.

关 键 词：三维面形测量 小波神经网络 函数逼近 傅里叶变换轮廓术 

分 类 号：TN247[电子电信—物理电子学] TP183[自动化与计算机技术—控制理论与控制工程]