基于平面靶分割区间LS-SVM摄像机标定的研究

Calibration of multi camera visual systems using divided LS-SVM

摄像机标定是视觉系统精确的前提.利用最小二乘支持向量机实现摄像机的标定不需确定摄像机具体的内部参数和外部参数.由于镜头的畸变主要由径向畸变引起,根据摄像机畸变的特点,对畸变区域进行划分,提出一种基于分割区间最小二乘支持向量机的摄像机标定法,对不同的畸变区域进行单独处理.该方法同BP神经网络和基本最小二乘支持向量机标定预测结果对比表明,基于分割区间最小二乘支持向量机的摄像机标定法速度快,实时性好,能有效提高标定精度.

Camera calibration is vital for any accurate visual system. This process can be simplified using least squares support vector machines （LS-SVMs） to achieve camera calibration. Internal and external parameters of the camera do not need to be analyzed. Because lens distortion is mostly caused by radial distortion. The distortion region was divided into zones according to the lenses distortion characteristics. A new method of camera calibration based on divided LS-SVMs was proposed where the distortion of different regions could be dealt with separately. Comparisons between results from a back-propagation （BP） neural network and the proposed LS-SVM show that calibration accuracy can be improved with LS-SVM processing zonal divisions of the image. This allows processing to be sufficiently quick and accurate that it can reliably provide real time results .