摘要
提出一种基于神经网络的系统标定方法.通过射影变换及误差补偿方法,建立摄像机图像平面与投影仪图像平面的映射关系,利用该映射关系和标定点的摄像机图像坐标,计算得到相应的投影仪图像坐标;建立三层结构的神经网络,该网络以两个图像坐标为输入,对应的世界坐标为输出,训练样本由得到的标定点的两个图像坐标及其世界坐标组成,采用BP算法训练该网络;训练过程即为神经网络逼近系统模型的过程,训练完成时,系统完成标定.实验表明,与传统的结构光标定方法对比,本文提出的方法简化了建模复杂度和标定过程,提高了标定精度,并具有普遍适应性.
A structured-light system calibration method based on a neural network was proposed. Byusing the method of projective transformation and error compensation, the mapped relation between thecamera image plane and the projector image plane was obtained. Then, with the relation and the cameraimage-coordinates, the corresponding projector image coordinates were calculated. So, a three-layerneural network was constructed. For this network, the inputs are two image coordinates and outputs are3D world coordinates. The training set consists of two image coordinates and 3D world coordinates ofcalibration points. Then, the neural network was trained by Back Propagation (BP) algorithm while thesystem model was fitting with it. When the process of the training was finished, the calibration was alsoaccomplished. The results of the experiments prove that the method proposed in the paper reveals ahigher degree of accuracy comparing with the conventional methods, and reduces the complexity of themodel and simplifies the process of calibration. Besides, it can be applied in various conditions gererally.
出处
《光子学报》
EI
CAS
CSCD
北大核心
2016年第5期81-86,共6页
Acta Photonica Sinica
基金
国家自然科学基金(No.61377104)资助~~
关键词
三维轮廓测量
结构光
摄像机标定
神经网络
BP算法
射影变换
模型
Surface measurement
Structured-light
Neural network
Cameras calibrationBackDroDa
ation, Projective transformation
Model