摘要
提出了一种新的基于Zernike矩和粒子群(PSO)算法的摄像机BP神经网络标定方法。首先,利用Zernike矩和曲率不变性求取圆形标定模板中心的亚像素坐标,提高神经网络训练数据的精度;其次,利用PSO算法优化网络的初始权重和阈值,提高网络的收敛速度和泛化能力。实验结果表明,该方法在X轴和Y轴方向的测量误差小于0.06 mm,整个测试集均方根误差为0.194 mm,证明了该方法的有效性。
A novel BP neural network on camera calibration approach is proposed based on Zernike moment and particle swarm optimization algorithm.First to improve the accuracy of training data,the subpixel coordinates of circular centers are detected by Zernike moment and curvature preserving.And secondly optimal weights and thresholds of neural networks are optimized by PSO algorithm to improve convergence rate and generalization ability.The experiments show that the errors are less than 0.06 mm in the X and Y axis direction and the root-mean-square error of test set is merely 0.194 mm,which indicate that the technique is feasible and effective.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2010年第9期1311-1314,共4页
Journal of Optoelectronics·Laser
基金
陕西省自然科学基金资助项目(2007E218)
陕西省教育厅自然科学专项资助项目(09JK559)