构建虚拟立体靶标的大视场高精度视觉标定

为了提高立体视觉系统在大视场下的测量精度,基于误差溯源思想提出了一种构建虚拟立体靶标的大视场高精度视觉系统标定方法,克服了大尺寸高精度标定物难以制造等问题。对影响立体视觉系统测量精度的主要因素进行分析,列出视觉测量系统的误差溯源链,解析了大视场视觉系统精度瓶颈的原因。借助激光跟踪仪,运用非线性最小二乘单位四元数算法求解坐标系刚体变换,获取大范围高精度的空间点阵,构建虚拟靶标。在相机畸变模型中考虑了三阶径向畸变和二阶切向畸变参数,并使用Levenberg-Marquardt迭代算法进行标定参数求解,进一步提高系统精度。实验构建了一套测量空间约为4m×3m×2m的双目立体视觉系统,通过对某型号高精度直线导轨进行点距测量,在测量距离3m处,152组不同长度的横向距离测量的误差算术均值为-0.003mm,误差标准差为0.08mm。测量精度相较于传统的平面标定法有较大提升。

High precision calibration of vision measurement system in large FOV based on virtual 3D target

In order to achieve high-precision calibration of camera with large Field of View (FOV), a stereo calibration method based on error tracing thought was proposed, which can overcome the manufacture limitation of targets with large size and high-accuracy in practice.The main factors affecting accuracy of stereo-vision system were analyzed and the error track chain of vision system was listed, which can explain the accuracy bottleneck in vision system with large FOV.Then, laser tracker was adopted to calculate the rigid body transformation with unit quaternion method, thus achieving virtual point clouds with high precision.Subsequently, 3-order radial distortion and 2-order tangential distortion were involved in the mathematic camera distortion model.In addition, the calibration parameters were solved through Levenberg-Marquardt iterative algorithm, which can improve the accuracy moderately.In the experiment, the binocular stereo vision measuring system was designed with FOV of 4 m×3 m×2 m.The results show that the reconstruction mean error for 152 group data of lateral distances is-0.003 mm and the standard deviation is 0.08 mm in the detection of a high-precision guide rail.Compared with the conventional plane calibration method, the measuring accuracy of the stereo calibration method has been greatly improved.