基于优化RBF-DDA神经网络的摄像机标定

Camera Calibration Based on Optimized RBF-DDA Neural Networks

设计了一种改进的基于动态衰减的RBF神经网络,它能够自适应地确定RBF隐层节点数、高斯函数中心值及径向基函数的宽度,克服了原算法中过多依赖先验知识设计参数的弊病,仿真实验验证了该算法的有效性。并且将此网络应用于摄像机定标中,该网络无需确定摄像机具体的内、外部参数,而且补偿了摄像机非线性畸变,使测量结果更加准确。实验结果表明,应用该神经网络进行摄像机标定能达到较高的精度,且在机器人平面跟踪实验中得到了令人满意的结果。

This paper presents an algorithm of dynamic decay adjustment RBF neural networks which can adaptively get the number of the hidden layer nodes, the center values of Gaussian function and the width of RBF. The neural networks overcome the difficulty of fixing parameters in the former neural networks. The experimental results show that this algorithm is effective. Then it is applied into camera calibration. It doesn＇t require an accurate mathematical model and compensates for the nonlinear distortion of camera, which makes the outcome more accurate. The experimental results show that this neural network calibration can obtain high accuracy and it is used in the robot curve tracking successfully.