Research on LiDAR Calibration Algorithm for Non-Common-View Areas

Abstract

In recent years, artificial intelligence has developed rapidly, and people’s demand for intelligent production has been steadily increasing. Autonomous driving has also gradually attracted widespread attention. At present, vehicles are capable of independently completing key tasks such as environmental perception, map construction, high-precision positioning, path planning and vehicle control. To ensure the adaptability of driverless vehicles to complex environments, autonomous vehicles often deploy dual Light Detection and Ranging (LiDAR) sensors, yet mounting constraints frequently result in non-overlapping fields of view, rendering traditional overlap-based calibration ineffective. This paper therefore presents a calibration algorithm tailored for such configurations. First, it introduces a differential-constraint time-synchronization module to align the point clouds, followed by a denoising-and-downsampling pipeline to boost efficiency. The core contribution is a self-calibration method grounded on geometric-consistency priors: the primary LiDAR constructs a high-precision global map, against which the auxiliary LiDAR data are iteratively registered via Generalized Iterative Closest Point (GICP) optimization. Experimental results show that in the experimental scenario of the AGV trolley, the LiDAR has no common view area. The maximum translation error obtained by this algorithm is 0.064m and the rotation error is 1.243°. The above experimental results meet the accuracy requirements of the system, and the method delivers a practical, low-cost solution for multi-LiDAR fusion in complex environments.