基于WNN的双目摄像机标定方法

Camera calibration of binocular vision system based on wavelet neural network

针对传统摄像机标定方法需要建立复杂的数学模型,且计算量大、实时性不好的问题,引入了人工神经网络来有效处理非线性映射问题,准确地建立起立体视觉中三维空间特征点与它在图像平面上像点之间的非线性关系;但现有的神经网络标定法仍存在实时性差、标定精度不够、泛化能力差的缺点,于是该文提出了一种基于小波神经网络(wavelet neural network,WNN)的方法,同时用粒子群优化算法对学习算法进行改进,并对小波网络与BP网络的标定结果进行比较.实验结果表明,基于小波神经网络的双目视觉标定方法能够达到较高的实时性、标定精度和泛化能力的要求.

For the problems of needing many complicated mathematical models with large amount of calculation and bad real-time performance in the traditional methods for camera calibration, the artificial neural networks were introduced to deal with the problem of non-linear mapping effectively, and create accurately the non-linear relationship between the three-dimensional feature point and its image point on the image plane. However, the existing calibration methods using neural networks still have the disadvantages of bad real time, poor calibration accuracy and generalization ability. So this paper proposed a method based on wavelet neural network, while using particle swarm optimization to improve learning algorithm, and compared the results of calibration with BP neural network method. Experimental results showed that camera calibration of binocular vision system based on wavelet neural network athieved a better real time, higher calibration accuracy and generalization capabilities