Enhancing Inter-frame Registration Accuracy Based on Adaptive Grid Projection

The high-precision positioning of multi-line LiDAR in urban environments relies on the accuracy of inter-frame point cloud registration. However, dynamic objects and changes in local geometric features in complex scenes can affect the stability of traditional registration methods. This paper proposes an inter-frame registration method based on adaptive grid projection to improve registration accuracy. The proposed method first preprocesses the point cloud by extracting feature points for registration and applying Euclidean clustering to generate point cloud clusters. Each corresponding feature cluster is labeled, and an adaptive grid projection based on the incident angle is employed to enhance pole-like features. Additionally, low-stability regions are removed to improve the reliability of matching points. Subsequently, inter-frame registration is performed using pole-like features to achieve high-precision pose estimation. Experimental results demonstrate that the proposed method effectively enhances registration accuracy and robustness in urban environments, providing reliable technical support for LiDAR-based autonomous localization in GNSS-denied scenarios.