基于神经网络的摄像机平面模板标定方法

Method of camera calibration with planar pattern based on neural network

根据立体视觉原理,提出了基于神经网络摄像机标定方法,该方法的特点是采用平面模板作为标定物。由于传统的标定算法在考虑各种非线性畸变时计算十分繁琐,利用BP神经网络来直接学习图像信息与三维信息之间的关系,可实现从输入到输出的任意非线性映射。为了提高BP神经网络训练速度和避免陷入局部最小解,采用引入动态因子的学习算法。实验结果表明,该算法具有较高的标定精度,而且可行。

According to the thery of stereo vision,a method of calibration based on neural network is proposed,which only requires a planar pattern.Because the conventional calibration algorithms taking account of nonlinear distortions are very complex,and BP neural networks are used to learn the relationships between the image information and the 3D information,and BP neural networks can realize any nonlinear relationship from input to output.A dynamic gene is introduced to the BP algorithm to improve the learning speed of the neural network and to avoid getting in local least.The experimental results show that the algorithm has high precision,and the algorithm is feasible.