Extrinsic calibration of multiple 2D Lidars using a Genetic Algorithm: A robust method to estimate the x, y and yaw coordinates of vehicle-mounted 2D Lidars using data acquired when moving

Abstract

To correctly transform and merge measurements from multiple vehicle-mounted Lidar sensors, it is essential to know the correct position of each Lidar. By using a genetic algorithm we have designed an alternative calibration technique which simultaneously estimates the x, y, and yaw coordinates of four 2D Lidars. A grid-based approach with decreasing cell size is used to evaluate the alignment of point cloud merging. Additionally, the method overcome difficulties of sparse input data by utilizing information from multiple perspectives attained as the vehicle is moving. With this feature the algorithm successfully calibrate the Lidar setup in more than 95% of the test cases. These tests are performed using data from indoor or outdoor environments and show that the technique is very much capable of calibrating such Lidar setups irregardless of the operating environment. Gathering data while the Lidars are moving induce distortions of the Lidar scans, wherefore compensation is also taken into account.