Boundary Net Extraction and Completion of Road Boundaries With Deep Learning Using Mobile Laser Scanning Point Clouds and Satellite Imagery
Abstract : Robust road boundary extraction and completion play an important role in providing guidance to all road users and supporting high-definition (HD) maps. The significant challenges remain in remarkable and accurate road boundary recovery from poor road boundary conditions. This paper presents a novel deep learning framework, named BoundaryNet, to extract and complete road boundaries by using both mobile laser scanning (MLS) point clouds and high-resolution satellite imagery. First, road boundaries are extracted by conducting a curb-based extraction method. Such extracted 3D road boundary lines are used as inputs to feed into a U-shaped network for erroneous boundary denoising. Then, a convolutional neural network (CNN) model is proposed to complete the road boundaries. Next, to achieve more complete and accurate road boundaries, a conditional deep convolutional generative adversarial network (c-DCGAN) with the assistance of road centerlines extracted from satellite images is developed. Finally, according to the completed road boundaries, the inherent road geometries are calculated. The proposed methods were evaluated using satellite imagery and four MLS point cloud datasets with varying densities and road conditions in urban environments. The quality evaluation metrics of 82.88%, 82.43%, 88.86%, and 84.89% were achieved for four data sets. The experimental results indicate that the BoundaryNet model can provide a promising solution for road boundary completion and road geometry estimation.
? A mobile laser scanning (MLS) system is developed as a cost-effective solution for large-scale 3-D data acquisition.
? Some methods also utilize existing geographic information system maps to extract roads from MLS point clouds.
? To create a training set for a U-Net-based neural network that is able to detect road markings, more than one thousand road markings in about 600 images have been marked using existing manually created shape files as ground truth data.
? We assume that the lowest points of road curbs constitute road boundaries that separate pedestrian pavements or other green spaces from road surfaces.
? We mainly focus on the theoretical and methodological problems of road boundary extraction and completion using MLS point clouds and satellite imagery.
? Thus, the complicated erroneous line removal problem is perceived as a straightforward binary image classification task.
? The dropout operation is implemented to reduce the over-fitting problem.
? To address this problem, multiple scans can be conducted at varying scan directions to collect more point clouds of roads, which remarkably improve the BoundaryNet model.
? Furthermore, different image resolutions have significant impacts on road boundary extraction.
• The performance of the proposed method is evaluated on the point clouds of an urban scene collected by a RIEGL VMX-450 MLS system.
• Some methods have been proposed to detect road boundaries from MLS point clouds.
• To extract the lowest points of curbs, a projection-based method is proposed in this letter.
• We have proposed a novel local-normalsaliency-based method to extract road boundaries from MLS point clouds.
• The proposed method directly constructs a saliency map on 3-D laser scanning point clouds and extracts road curbs according to their spatial location relationship with contextual objects.
? This method indicated superior performance in line drawing restoration and curvature and thickness conservation.
? The proposed c-DCGAN model delivers better completion performance on the missing parts of smaller sizes.
? Although the completeness and curvatures vary in these road scenarios, the experimental results indicate that the BoundaryNet obtains good road boundary completion performance.
? Therefore, the BoundaryNet achieves superior performance in road boundary completion, which can deal with incomplete road boundaries with varying completeness rates.
? To further demonstrate the robustness and efficiency of the Boundary Net in relatively low-quality point clouds, we evaluated the Boundary Net performance using the Paris-Lille-3D dataset.
|