UNORGANIZED-LIDAR POINT CLOUD-DATASET

Pavement planar coefficients are critical for a wide range of civil engineering applications, including 3D city modeling, extraction of pavement design parameters, and assessment of pavement conditions. Existing plane fitting methods, however, often struggle to maintain accuracy and stability in complex road environments, particularly when the point cloud is affected by non-pavement objects such as trees, curbstones, pedestrians, and vehicles.

REoPC is proposed as a robust two-stage estimation method based on road point clouds acquired using a hybrid solid-state LiDAR. The method consists of two main parts: coarse estimation and refined estimation. The first stage employs a dual-plane sliding window to remove major outliers and extract an initial surface. The second stage introduces a new cost function based on the Geman-McClure estimator to further suppress residual noise and reduce fitting instability caused by outlier influence and algorithmic randomness.

Evaluation is conducted on both synthetic and real-world datasets collected using a custom mobile LiDAR scanning system across three urban road scenarios—flat, crowned, and traffic-interfered segments. Each scenario includes 100 frames of road point clouds, with approximately 35,000 points per frame, offering a diverse and challenging benchmark. REoPC consistently outperforms existing methods in terms of accuracy and robustness and exhibits low sensitivity to parameter tuning, demonstrating strong applicability in varied real-world conditions.