LiDAR and Camera Online Calibration Based on a Semantic Reliable Centroid Algorithm

A novel method for online calibration of cameras and LiDAR extrinsic parameters is proposed. Previous online calibration methods first required offline calibration followed by fine-tuning or optimization online. The proposed online calibration method uses instance segmentation and semantic segmentation to identify common features of camera and LiDAR images and then calculates the centroids of these features to correct the paired points to establish the initial extrinsic parameters. Previous offline calibration is not required. In addition, the problems of angle and point cloud inhomogeneity were considered; the algorithm selects either the contour method or the cluster method to calculate the centroid to maximize the accuracy of the initially calculated extrinsic parameters. These parameters are then optimized using a designed cost function; minimizing the cost function produces the selected extrinsic parameters as the calibration result. The method proposed in this study was evaluated on the Kitti Dataset. The experimental results reveal that the method in this study is feasible in terms of both its initialization strategy and its online calibration accuracy for both the LiDAR and the camera.