Lane-Level Road Extraction from High-Resolution Optical Satellite Images

Abstract

High-quality updates of road information play an important role in smart city planning, sustainable urban expansion, vehicle management, urban planning, traffic navigation, public health and other fields. However, due to interference from road geometry and texture noise, it is difficult to avoid the decline of automation while accurately extracting roads. Therefore, we propose a high-resolution optical satellite image lane-level road extraction method. First, from the perspective of template matching and considering road characteristics and relevant semantic relations, an adaptive correction model, an MLSOH (multi-scale line segment orientation histogram) descriptor, a sector descriptor, and a multiangle beamlet descriptor are proposed to solve the interference from geometry and texture noise in road template matching and tracking. Second, based on refined lane-level tracking, single-lane and double-lane road-tracking modes are designed to extract single-lane and double-lane roads, respectively. In this paper, Pleiades satellite and GF-2 images are selected to set up different scenarios for urban and rural areas. Experiments are carried out on the phenomena that restrict road extraction, such as tree occlusion, building shadow occlusion, road bending, and road boundary blurring. Compared with other methods, the proposed method not only ensures the accuracy of lane-level road extraction but also greatly improves the automation of road extraction.

A Model-Driven-to-Sample-Driven Method for Rural Road Extraction

“…In the tracking process, local direction prediction, multicircle template construction, and panchromatic and HSV space interactive matching were started with a step size of twice the road width, and the extraction of rural roads was completed by iterative tracking. In addition, if an occlusion did not meet the requirements in the tracking process, the step size could be increased to 5 times the road width [40].…”

“…(1) Prediction of local road direction Considering the irregular curvature changes in rural roads, changes in the local road direction can be greater than 120 degrees; if no local road direction prediction is included, the matching point is readily placed outside of the road, which inevitably increases the degree of manual participation. The MLSOH descriptor [40] establishes a road local direction prediction model based on the semantic relationship between the local road direction and the edge direction of neighboring objects and the direction and length of the local range detection line segment. However, in our method, considering the irregular curvature changes in rural roads, it is difficult to ensure the accuracy of direction prediction with the length of a single line segment.…”

“…Therefore, we chose the template matching method that could form vector results. These three methods are the T-shaped template of Lin Xiangguo et al [38], the circular template of Lian Renbao et al [39], and the sector descriptor method proposed by Dai Jiguang et al [40]. The T-shape model uses the angle texture feature to accurately locate the initial road point and calculate the width of the road and the direction of the road and uses gray least square matching to extract the center point of the road.…”

“…In this method, local effective samples are formed through manual input of initial information, the parameter information of local samples is learned as the benchmark of matching and tracking, and roads are extracted by human-machine interaction. Classic templates include section templates [36], rectangular templates [37], T-shaped templates [38], circular templates [39], and sector descriptors [40,41]. For example, Zhang et al [42] manually input three points on the edges of both sides of a road, constructed a rectangular template and input direction parameter information, and completed the extraction of the road through an iterative method.…”

Automatic Road Extraction from High-Resolution Remote Sensing Images Using a Method Based on Densely Connected Spatial Feature-Enhanced Pyramid

“…The methods for remote sensing image road extraction are mainly divided into feature-based, classification-based and deep learning approaches [4]- [5]. Feature-based methods mainly extract roads from images by considering their features, such as road extraction methods based on shape [6], texture [7]- [9], and geometry [10], [11]. Feature-based methods have a good effect on simple and regular road extraction, but they have poor extraction effects on complex roads and require substantial postprocessing to repair the initially extracted roads.…”