神经网络获取三维面形研究  

BP neural network applied to 3D object measurement based on fringe pattern projection

作　　者：唐燕[1] 陈文静[1] 张强[1] 苏显渝[1] 向立群[1] 

机构地区：[1]四川大学光电子科学技术系,四川成都610064

出　　处：《光电工程》2007年第12期61-65,共5页Opto-Electronic Engineering

基　　金：国家自然科学资金资助(60677028)

摘　　要：针对基于传统三角原理的结构光三维测量方法难以测量阶梯形物体的问题,应用神经网络对获取的变形条纹进行处理,获取物体的三维面形信息。该方法通过对神经网络的训练,直接建立条纹图分布与物体高度之间的对应关系,完成对物体的三维测量,即使在投影系统参数未知的情况下,也能取得较好的结果。论文中提出的神经网络三维面形测量方法测量时间短,测量过程中只需要一幅条纹图就能恢复阶梯物体的高度信息。计算机模拟及试验验证了方法的可行性。In order to,neasure the stepped shape object which can＇t be recovered by the traditional triangle method, a neural network is used to deal with this kind of fringe patterns for obtaining the three-dimensional （3D） information of the object. By training the network, the relationship between the fringe pattern and the height of the object can be established, therefore, the height of the object can be obtained. Furthermore, the object still can be reconstructed perfectly without knowing the optical parameters of the experiment system. An obvious merit of the method is that it can recover the 3D object in a short time and need only one fkinge pattern. Computer simulations and experiment validate the feasibility of the method.

关 键 词：三维面形测量 BP神经网络 条纹图 

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]