作　　者： ; ; ; ;

机构地区： 吉林大学汽车工程学院汽车动态模拟国家重点实验室

出　　处： 《农业机械学报》 2009年第8期214-218,共5页

摘　　要： 以前馈型BP神经网络进行双目立体视觉系统的摄像机标定研究,基于Harris角点提取,提出了增加约束的改进方法,从而提高网络训练样本集的数据精度;探讨了神经网络的欠泛化、过泛化问题,综合运用归一化、提前终止等多种策略,进一步提高网络泛化能力,并与经典标定方法进行对比。试验结果表明,该方法能够获得较高的摄像机标定精度。 Camera calibration in a binocular stereo vision system was studied based on BP neural network techniques. A neural network was built to investigate the relationship between the image coordinates and the space coordinates. To improve the accuracy of training data, a corner extraction algorithm based on Harris algorithm was modified by increasing constraints. The generation ability of neural networks was discussed, and was further improved by synthesizing many strategies such as the regularization and stopped training strategies. At last, compared with the traditional calibration method, the test result shows that this method is available and can get a higher precision for binocular camera calibration.

关 键 词： 摄像机标定 神经网络 角点 泛化

领　　域： [自动化与计算机技术] [自动化与计算机技术]