Accurate and efficient extraction of road marking plays an important role in road transportation engineering, automotive vision, and automatic driving. In this article, we proposed a dense feature pyramid network (DFPN)-based deep learning model, by considering the particularity and complexity of road marking. The DFPN concatenated its shallow feature channels with deep feature channels so that the shallow feature maps with high resolution and abundant image details can utilize the deep features. Thus, the DFPN can learn hierarchical deep detailed features. The designed deep learning model was trained end to end for road marking instance extraction with mobile laser scanning (MLS) point clouds. Then, we introduced the focal loss function into the optimization of deep learning model in road marking segmentation part, to pay more attention to the hard-classified samples with a large extent of background. In the experiments, our method can achieve better results than state-of-the-art methods on instance segmentation of road markings, which illustrated the advantage of the proposed method.

The three-dimensional (3D) geographic locations of street furniture such as traffic signs comprise the basic content of 3D city construction, and such information is indispensable for periodic statistics for road management and maintenance. This paper presents a novel solution for acquiring 3D information on small traffic signs based on Mobile Mapping System (MMS) data. Firstly, a lightweight backbone network called VGG-L under an optimized Faster Region-based Convolutional Neural Network (R-CNN) detection framework is proposed for the detection of small traffic signs. An Urban Traffic Sign Detection (UTSD) dataset is created based on panoramic images obtained from test areas. Detection results from the UTSD dataset show that VGG-L outperforms other popular networks and achieves a mean average precision of 75.4%, which is 4.2—14.8% higher than those of VGG16, MobileNet, ResNet, and YOLOv3. Secondly, a novel depth-map-based 3D spatial geo-location method is proposed for obtaining the 3D geographic locations of the objects. Then, a center-based method is proposed to automatically extract the final 3D vector of the target. Experimental results illustrate that the proposed method performs 3D positioning and vectorization of the milestones, circular and triangular traffic signs, accurately and effectively, achieving greater than 86% recall and precision for the three types of targets in the test areas. The experiments demonstrate that the overall 3D information acquisition scheme is feasible and has great application potential.

Road markings are one of the most important safety elements in a road network, and they play a critical role in traffic safety. However, the automatic extraction of road markings remains a technical challenge in the fields of smart city construction and automatic driving. This paper presents an image-translation-based method of obtaining the 3D vectors of typical road markings from mobile laser point clouds. First, ground roughness is used as a criterion to extract ground points based on the topological relationship of adjacent scan lines, and the feature images of a road surface are generated using the adapted inverse distance weighted method. Second, by comparing objective functions based on the pix2pix framework, a finely adjusted image-to-image translation model named P2P_L1 is proposed for the segmentation of road markings. The proposed model outperforms the advanced DeepLab V3+ network in terms of precision, F1-score, and mean Intersection over Union indicators in the comparative segmentation results of ten types of road markings in the Shenzhen test area. Third, methods such as node averaging and optimized iterative closest point are developed for the 3D vectorization of road markings. This study presents a new approach for the automatic extraction of road markings to provide effective technical support for the construction of smart cities.

The GF-7 satellite is China’s first high-resolution stereo mapping satellite that reaches sub-meter resolution, equipped with new-type payloads, such as an area array footprint camera that can achieve synchronization acquisition of laser spots. When the satellite is in space, the variation of camera parameters may occur due to launch vibration and environmental changes, and on-orbit geometric calibration thereby must be made. Coupled with the data from the GF-7 satellite, this paper constructs a geometric imaging model of the area array footprint camera based on the two-dimensional direction angle, and proposes a coarse-to-fine “LPM-SIFT + Phase correlation” matching strategy for the automatic extraction of calibration control points. The single-image calibration experiment shows that the on-orbit geometric calibration model of the footprint camera constructed in this paper is correct and effective. The matching method proposed is used to register the footprint images with the DOM (Digital Orthophoto Map) reference data to obtain dense control points. Compared with the calibration result using a small number of manually collected control points, the root mean square error (RMSE) of the residual of the control points is improved from half a pixel to 1/3, and the RMSE of the same orbit checkpoints in the image space is improved from 1 pixel to 0.7. It can be concluded that using the coarse-to-fine image matching method proposed in this paper to extract control points can significantly improve the on-orbit calibration accuracy of the footprint camera on the GF-7 satellite.

The correction of spaceborne laser footprint positioning errors in forested areas is the primary task required to improve the reliability of laser data and ensure the accuracy of large-scale forestry surveys based on laser data. Currently, this correction largely requires high-precision terrain data, which are difficult to obtain over mountainous forests. Therefore, a method for correcting spaceborne laser footprint positioning errors in mountainous forest areas by minimizing the sum of satellite‒Earth ranging residuals is proposed. In this method, a satellite digital surface model (DSM) is used to calculate satellite‒Earth laser ranges at different ground locations, and the laser footprint positioning error is corrected based on the principle of minimizing the sum of the ranging residual between the satellite‒Earth laser ranging and the actual ranging. The proposed method was subsequently tested on the GF7-01 satellite laser, and the corrected laser footprint positions were compared with the true positions of the laser footprints captured by laser detectors. The results indicated that the laser footprint positioning accuracy was 4.16 m for beam 1 of the GF7-01 satellite, while beam 2 achieved an accuracy greater than 3.96 m. Overall, these outcomes clearly demonstrate the effectiveness of the proposed method.