Mobile Mapping Technology (MMT) is one of the most important 3D spatial data acquisition technologies. The state-of-the-art mobile mapping systems, equipped with laser scanners and named Mobile LiDAR Scanning (MLS) systems, have been widely used in a variety of areas, especially in road mapping and road inventory. With the commercialization of Advanced Driving Assistance Systems (ADASs) and self-driving technology, there will be a great demand for lane-level detailed 3D maps, and MLS is the most promising technology to generate such lane-level detailed 3D maps. Road markings and road edges are necessary information in creating such lane-level detailed 3D maps. This paper proposes a scan line based method to extract road markings from mobile LiDAR point clouds in three steps: (1) preprocessing; (2) road points extraction; (3) road markings extraction and refinement. In preprocessing step, the isolated LiDAR points in the air are removed from the LiDAR point clouds and the point clouds are organized into scan lines. In the road points extraction step, seed road points are first extracted by Height Difference (HD) between trajectory data and road surface, then full road points are extracted from the point clouds by moving least squares line fitting. In the road markings extraction and refinement step, the intensity values of road points in a scan line are first smoothed by a dynamic window median filter to suppress intensity noises, then road markings are extracted by Edge Detection and Edge Constraint (EDEC) method, and the Fake Road Marking Points (FRMPs) are eliminated from the detected road markings by segment and dimensionality feature-based refinement. The performance of the proposed method is evaluated by three data samples and the experiment results indicate that road points are well extracted from MLS data and road markings are well extracted from road points by the applied method. A quantitative study shows that the proposed method achieves an average completeness, correctness, and F-measure of 0.96, 0.93, and 0.94, respectively. The time complexity analysis shows that the scan line based road markings extraction method proposed in this paper provides a promising alternative for offline road markings extraction from MLS data.

Mobile LiDAR Scanning (MLS) systems and UAV LiDAR Scanning (ULS) systems equipped with precise Global Navigation Satellite System (GNSS)/Inertial Measurement Unit (IMU) positioning units and LiDAR sensors are used at an increasing rate for the acquisition of high density and high accuracy point clouds because of their safety and efficiency. Without careful calibration of the boresight angles of the MLS systems and ULS systems, the accuracy of data acquired would degrade severely. This paper proposes an automatic boresight self-calibration method for the MLS systems and ULS systems using acquired multi-strip point clouds. The boresight angles of MLS systems and ULS systems are expressed in the direct geo-referencing equation and corrected by minimizing the misalignments between points scanned from different directions and different strips. Two datasets scanned by MLS systems and two datasets scanned by ULS systems were used to verify the proposed boresight calibration method. The experimental results show that the root mean square errors (RMSE) of misalignments between point correspondences of the four datasets after boresight calibration are 2.1 cm, 3.4 cm, 5.4 cm, and 6.1 cm, respectively, which are reduced by 59.6%, 75.4%, 78.0%, and 94.8% compared with those before boresight calibration.

Registration of point clouds is a fundamental issue in Light Detection and Ranging (LiDAR) remote sensing because point clouds scanned from multiple scan stations or by different platforms need to be transformed to a uniform coordinate reference frame. This paper proposes an efficient registration method based on genetic algorithm (GA) for automatic alignment of two terrestrial LiDAR scanning (TLS) point clouds (TLS-TLS point clouds) and alignment between TLS and mobile LiDAR scanning (MLS) point clouds (TLS-MLS point clouds). The scanning station position acquired by the TLS built-in GPS and the quasi-horizontal orientation of the LiDAR sensor in data acquisition are used as constraints to narrow the search space in GA. A new fitness function to evaluate the solutions for GA, named as Normalized Sum of Matching Scores, is proposed for accurate registration. Our method is divided into five steps: selection of matching points, initialization of population, transformation of matching points, calculation of fitness values, and genetic operation. The method is verified using a TLS-TLS data set and a TLS-MLS data set. The experimental results indicate that the RMSE of registration of TLS-TLS point clouds is 3~5 mm, and that of TLS-MLS point clouds is 2~4 cm. The registration integrating the existing well-known ICP with GA is further proposed to accelerate the optimization and its optimizing time decreases by about 50%.

Point cloud filtering is an essential preprocessing step in 3D (three-dimensional) LiDAR (light detection and ranging) point cloud processing. The filtering of mobile LiDAR scanning point clouds is much more challenging due to their non-uniform distribution, the large-scale of missing data areas and the existence of both large size objects and small land features. This paper proposes a new filtering method that combines range constraint, slope constraint and angular position constraint to filter ground surface points from mobile LiDAR point clouds. Firstly, a cylindrical coordinate system (CCS) is established for each block of mobile LiDAR point clouds and three attributes of mobile LiDAR points, i.e., the angular position attribute (AA), longitudinal distance attribute (LA) and range attribute (RA), are computed. Then, the mobile LiDAR point clouds are structured into a grid according to the AA and LA. Finally, the point clouds are filtered by a single cross-section filter (SCSF) using range constraint and slope constraint, followed by a multiple cross-section filter (MCSF) using range constraint and angular position constraint. Five datasets are used to validate the proposed method. The experimental results show that the proposed new filtering method achieves an average type I error, type II error, and total error of 1.426%, 1.885%, and 1.622%, respectively.

Deep learning methods have been widely used to map building contours automatically in high-resolution remote sensing images over recent years. However, most deep learning-based methods still require complex post-processing to generate regular contours. P2PFormer is an advanced method that can directly obtain the positions and sequences of general geometric primitives like points, lines, and angles (corners) of a building instance without complex post-processing. Nevertheless, due to the inherent limitations of ROI-Align, P2PFormer faces significant challenges in primitive detection. ROI-Align utilizes a uniform sampling approach for feature extraction, it inevitably introduces a high proportion of invalid sampling points into the extracted features. Resulting in missed and false detections of building primitives, ultimately affecting the accuracy of building contour extraction. To address these issues, we propose P2PFormerV2, which introduces a contour feature enhancer to improve P2PFormer. The contour feature enhancer increases the proportion of valid feature sampling points and enhances the extraction of contour-aware features, significantly improving the accuracy of primitive segmentation. This enhancer comprises three key components: the sparse feature extractor, the dense feature extractor, and the feature fusion module. The sparse feature extractor optimizes the sampling strategy to increase the proportion of valid feature sampling points; the dense feature extractor generates rich contour features and provides additional supervision signals; the feature fusion module integrates the outputs of the first two components to further enhance the extraction of contour features. Experimental results demonstrate that P2PFormerV2 achieves average precisions (AP) of 74.7%, 79.6%, and 64.2% in the WHU, CrowdAI, and WHU-Mix datasets, respectively, significantly outperforming the original P2PFormer and other existing advanced methods. Our findings about the shortcomings of ROI-Align and the importance of improving the effective feature extraction provides insights for future building extraction research.

Road intersections serve as critical topological nodes in transportation networks, and their rapid, accurate identification is essential for intelligent transportation systems and geographic information updates. However, direct detection of road intersections on large-scale high-resolution remote sensing imagery remains challenging due to intersection structural diversity, complex illumination conditions, and vegetation occlusion, which make it difficult to balance detection accuracy and computational efficiency. To address these challenges, Swin Transformer-GSConv-efficient IoU-YOLO (SW-GSE-YOLO), a novel lightweight and high-precision road intersection detection model, is proposed. The model incorporates three key innovations: 1) a Swin-C2PSA module that is designed by integrating the shifted-window self-attention mechanism of the Swin Transformer with the C2PSA structure, enabling efficient large-scale feature modeling while enhancing local detail and global contextual representation; 2) a GS3k2 lightweight network architecture is built upon GSConv and CSPNet, which reduces the parameter count while preserving feature extraction capability; and 3) the efficient intersection over union (EIoU) loss function is adopted to optimize bounding box regression and improve intersection center localization accuracy. To validate the proposed method, the RIS-42k dataset is constructed, comprising 42 372 annotated image patches of size $400\times 400$ pixels. Experimental results demonstrate that SW-GSE-YOLO outperforms recent mainstream lightweight detectors, including You Only Look Once (YOLO)v8–YOLOv12 and real-time DEtection TRansformer (RT-DETR). Compared with the YOLO11n baseline, it achieves improvements of 5.9%, 1.3%, 3.9%, and 14.6% in precision, recall, mAP at 0.5, and mAP at [0.5:0.95], respectively, while significantly reducing GPU memory usage and inference latency for large-scale image processing. Furthermore, SW-GSE-YOLO demonstrates robust localization accuracy and strong generalization capability across complex urban, suburban, and rural environments.