智能汽车激光雷达和相机数据融合系统标定

Calibration of lidar-camera fusion system for intelligent vehicles

智能汽车采用激光雷达和相机的数据融合实现对环境的感知,针对数据融合中不同传感器坐标系的联合标定问题提出了特征点法和棋盘格法两种标定方法。特征点法采用专门设计的标定模板,提取若干对激光雷达和图像对应点,建立约束方程组,采用最小二乘法求解结果。棋盘格法采用张正友标定法获取相机的内部参数,利用棋盘格平面在两个坐标系的一致性,建立约束方程组,采用线性方法求解两个坐标系的外部参数初始解,再用非线性优化方法进一步优化。利用两种方法得到的标定结果将激光雷达点投影到图像上并比较其对准精度。实验表明,两种方法都可以获取各传感器坐标系之间的位置关系,其投影对准误差分别为特征点法3.03像素和棋盘格法2.33像素。

Intelligent vehicles use a lidar-camera sensor fusion system to perceive the environment.Two calibration methods, feature point method and checkerboard method, are proposed for the joint calibration of different sensor coordinate systems in the data fusion. The feature-point method employs a tailored calibration template to extract several pairs of corresponding points, and solves the constraint equations for the calibration parameters in virtue of the least square method. The checkerboard method employs Zhang’s calibration method to obtain the intrinsic parameters of the camera. And then, equations are derived by using the consistency of the checkerboard plane in lidar and camera coordinate systems, solving the extrinsic parameters between the two coordinates using a linear method. The result is further refined by a nonlinear optimization method. The lidar points are projected onto the image plane by using the calibration results obtained from the two methods.Experiments demonstrate that the two methods are capable of obtaining the accurate position parameters between the coordinate systems of each sensor. The projection alignment errors are 3.03 pixels for the feature point method and 2.33 pixels for the checkerboard method.