一种鲁棒的LiDAR-IMU联合标定方法
A Robust LiDAR-IMU Joint Calibration Method
-
摘要: 针对目前主流的LiDAR-IMU联合标定方法在遮挡较为严重或者缺乏大块平面的复杂环境下标定精度较低的问题, 提出了一种鲁棒的LiDAR-IMU联合标定方法。首先, 在匹配构建阶段引入了定位精度高、不易受遮挡等影响的线特征, 并同时构建线特征和面片匹配对来增强标定约束; 其次, 在迭代优化阶段设计了一种双阶段的优化方法, 并根据每轮迭代优化的几何残差设计了自适应损失权重, 使得迭代优化过程能很好地收敛, 并提高了标定方法的精度。利用自建的室内数据集和开源的室外数据集对该方法进行了测试, 结果表明, 本文方法对于平移外部参数的标定标准差约为2 mm, 旋转外部参数的标定标准差约为0.04?, 优于当前主流标定方法的结果。Abstract: Current mainstream LiDAR-IMU (inertial measurement unit) joint calibration methods are of low accuracy in complex environments where there are serious occlusions or insufficient large planar surfaces. Facing that problem, a robust LiDAR-IMU joint calibration method is proposed. Firstly, line features are introduced in the matching construction stage, for they are not susceptive to occlusions and are of high localization accuracy. Both line feature matching pairs and planar patch matching pairs are constructed to strengthen the calibration constraint. Secondly, a two-stage optimization pipeline is constructed in the iterative optimization stage, where adaptive loss weights are designed according to the geometric residuals for each round of iterative optimization. Thus excellent convergence can be achieved by the optimization process, and the accuracy of the calibration method is improved. The proposed method is tested with the open-source outdoor dataset and the self-built indoor dataset. The results show that the calibration standard deviation of the proposed method for translation external parameters is about 2 mm and the calibration standard deviation for rotating external parameters is about 0.04?, which is much better than the state-of-the-art methods.