针对传统摄像机标定过程复杂、三维测量精度不高的问题, 提出了一种基于投影直线相交的双目视觉三维测量方法,并给出了基于神经网络的标定方法。根据摄像机成像特点, 利用双标定平面上的点求取投影直线方程; 针对摄像机成像复杂的畸变模型, 利用BP神经网络对复杂非线性映射关系的强大逼近能力, 对左右两台摄像机于远近标定平面处分别进行隐式标定, 得到图像像素坐标到平面模板二维物理坐标的映射关系; 设计制作了试验平台, 采用手工辅助的方法获得网格模板训练样本。此标定方法完全适用于大视场、近距离、高精度的双目视觉传感器。试验结果表明, 系统对空间已知长度的测量结果误差约为0.109 mm, 测量精度较高。

Considering the high complexity of traditional camera calibration process and the low accuracy of three-dimensional measurement we proposed a kind of binocular visual 3D measurement method based on projection line and a calibration method based on Neural Network.According to the characteristics of camera imaging the projection straight line equation was calculated by two points on the calibration plane.To the complex distorted model of camera imaging the powerful ability of BP neural network for fitting the complicated nonlinear mapping relation was used for left and right cameras implicit calibration at the far and near distance calibration planes and mapping relationship between two-dimensional physical coordinates of plane template and pixel coordinates of image was obtained.A test platform was designed and established, and grid template training samples were obtained by using the manual assisted method.This calibration method is fully applicable to the large field-of-view close range high-precision binocular vision sensor.Test results show that the measuring error of known length in the space is about 0.109 mm and the system has a high measuring accuracy.