Proposal on Implementation of Geometric Properties on Reference Points Reconstruction for Extrinsic Parameters Calibration of Sparse Point Cloud

In autonomous driving schemes, various sensors are required to be implemented to acquire various kinds of data. A calibration method is needed to find the relationship between two or more types of sensors to link them. Traditionally, a uniquely configured board is necessitated for LiDAR camera calibration. However, the traditional board contains excessive space. This study aims to minimize the required space for constructing the calibration board. Focusing on the properties of geometric shapes, the reconstruction of each circular shape from a hole can specify only a single reference point. Four circular holes with enough distance are required on a board to acquire four points in a row. If other geometrical holes are instead utilized, more than a single reference point can be extracted from a hole. In this experiment, proposed patterns are constructed based on two simple geometrical shapes, an equilateral, and a square. Compared with the conventional board, the points extracted from the proposed board with a triangular hole at the specific rotation achieved the highest accuracy while the size of the pattern was minimized.