Leveraging Crowdsourced GPS Data for Road Extraction From Aerial Imagery

Sun, Tao; Di, Zonglin; Che, Pengyu; Liu, Chun; Yin, Wang

doi:10.1109/cvpr.2019.00769

Cited by 83 publications

(62 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The road extraction from satellite images is no exception. Various deep models have been proposed for this task such as the restricted Boltzmann machines [25], FCN with UNet [39] and D-LinkNet [32]. A few works also leverage a generative model.…”

Section: B Satellite Image Applicationsmentioning

confidence: 99%

An Enhanced GAN Model for Automatic Satellite-to-Map Image Conversion

et al. 2020

View full text Add to dashboard Cite

Location-based service significantly relies on accurate and up-to-date maps. The conventional map generation involves labor-intensive and time-consuming manual efforts, which restricts the map-update frequency to a few years or even longer. In recent years, satellite images become more ubiquitous, and converting them to map-style images has attracted attention due to its frequent-updating and cost-effective nature. Generative adversarial network (GAN) is a promising approach for automatic satellite-to-map image conversion. However, it is still challenging to process satellite images when the underlying road structure is complex and irregular, or when some objects are visually indistinguishable due to obstruction or bad weather. To address these issues, we propose an enhanced GAN model to generate improved quality map images by bringing in the external geographic data as implicit guidance. The textual geographic data is converted to an image so that it can work collaboratively and seamlessly with the satellite image during the conversion. A high-level semantic regulation is also introduced to further reduce the noisy patterns generated during the translation, which occur frequently for the regions with sparse geographic data. The proposed method is versatile to various backbone GAN structures with a 20% performance improvement on three popular metrics (Inception Score, Frechet Inception Distance Score and SSIM score). Our proposed Semantic-regulated Geographic GAN (SG-GAN) is anticipated to reduce the manual identification efforts in broad geospatial applications.

show abstract

Section: B Satellite Image Applicationsmentioning

confidence: 99%

An Enhanced GAN Model for Automatic Satellite-to-Map Image Conversion

et al. 2020

View full text Add to dashboard Cite

show abstract

“…The D-Linknet consists of encoder-decoder structure, dilated convolution, and pre-trained encoder. Sun et al proposed a road extraction method using crowd-sourced GPS data to improve and support road extraction from aerial imagery [42].…”

Section: Road Extractionmentioning

confidence: 99%

Road Extraction from Very High Resolution Images Using Weakly labeled OpenStreetMap Centerline

Chen

et al. 2019

IJGI

View full text Add to dashboard Cite

Road networks play a significant role in modern city management. It is necessary to continually extract current road structure, as it changes rapidly with the development of the city. Due to the success of semantic segmentation based on deep learning in the application of computer vision, extracting road networks from VHR (Very High Resolution) imagery becomes a method of updating geographic databases. The major shortcoming of deep learning methods for road networks extraction is that they need a massive amount of high quality pixel-wise training datasets, which is hard to obtain. Meanwhile, a large amount of different types of VGI (volunteer geographic information) data including road centerline has been accumulated in the past few decades. However, most road centerlines in VGI data lack precise width information and, therefore, cannot be directly applied to conventional supervised deep learning models. In this paper, we propose a novel weakly supervised method to extract road networks from VHR images using only the OSM (OpenStreetMap) road centerline as training data instead of high quality pixel-wise road width label. Large amounts of paired Google Earth images and OSM data are used to validate the approach. The results show that the proposed method can extract road networks from the VHR images both accurately and effectively without using pixel-wise road training data. scribble annotation for road extraction. The ground truth of the pixel-level annotation in Figure 1b should label every pixel, which is difficult to generate. Figure 1c represents the scribble annotation, which can be easily obtained from OSM. Because the full annotation dataset is expensive to obtain and the scribble annotation is easy to generate, the study of road networks extraction using scribble labels is of great importance.Recently, weakly supervised learning was popular in image segmentation [13][14][15][16]. In these methods, scribbles [17,18], bounding boxes [19,20], clicks [19], and image-level tags [18] are used as supervision for image segmentation. In this work, the OSM centerline is used as a typical scribbles supervision for road extraction.In order to improve annotation efficiency and road extraction performance for automated VHR (Very High Resolution, spatial resolution 0∼2 m/pixel) images interpretation, this paper proposes a weakly supervised method to extract the road network supervised only by the scribble annotation OSM centerline. In this method, graph cut theory and a deep learning technique named Multi-Dilated-ResUNet (MD-ResUNet) are used to make efficient roads extraction.The main contributions of this paper are as follows:1. A novel deep learning approach based on revised ResUNet with hybrid loss is proposed for road extraction, which can only be supervised by weakly labeled OSM centerline instead of carefully notated pixel-wise road width information. 2. In order to improve the performance of the proposed model furtherly, a novel multi-dilation network with learnable parameters is added to conventional ResUnet. Th...

show abstract

“…Sun et.al. [9] incorporated the rendered GPS data as a new input of DCNN to help with aerial imagery for road extraction. However, this method just used one GPS trajectory feature and did not fully mine the features contained in trajectory.…”

Section: Extracting Road By Combining Rs Images With Trajectoriesmentioning

confidence: 99%

“…Simply using singe data source limits the performance of road recognition. We believed that trajectory data and remote sensing image can be combined to improve road continuity [9] and accuracy. Fig.1 indicates that these two data sources can enhance each other.…”

Section: Introductionmentioning

confidence: 99%

Fusing Taxi Trajectories and RS Images to Build Road Map via DCNN

Xiang

Cai-li

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Accurate road maps are fundamental to a wide range of applications, such as navigation, transportation and urban planning. With the rapid development of Globe Position System (GPS) and Remote Sensing (RS), abundant and continuously updated spatiotemporal data are available for road map building. However, using single data source inevitably results in limited performance on road information recognition, due to inherent defects in data, such as the noise, low sampling rate and uneven density distribution in trajectories, and occluded roads in RS-images. This paper aims to integrate GPS trajectories and RS-images to build road map, which is composed of intersections and links. First, both dynamic and static trajectory characteristics are utilized to capture road features indicated in trajectories, and meanwhile, road features implied in by RS-images are extracted via transfer-learning with an advanced deep neural network. Then, the two distinct types of road features are explored together to learn the potential area of road centerlines based on a variation of U-Net, which refines the U-Net to incorporate the structure characters of road. Finally, a multistep refinement is introduced to extract road centerlines through morphology processing, then derive intersections through connectivity analysis and detect links based on sliding-window. In addition, inspired by vector rasterization, a fast and automatic strategy is developed to create large-scale road dataset to significantly reduce labor and time. Compared with three state-of-the-art segment networks, the proposed network achieves the highest correctness and quality. The experimental results demonstrate that trajectory data and RS image complement each other, which assure the extracted road network possesses both integrity and connectivity. INDEX TERMS Road map generation, crowdsourcing trajectory data, remote sensing image, data fusion.

show abstract

Leveraging Crowdsourced GPS Data for Road Extraction From Aerial Imagery

Cited by 83 publications

References 28 publications

An Enhanced GAN Model for Automatic Satellite-to-Map Image Conversion

An Enhanced GAN Model for Automatic Satellite-to-Map Image Conversion

Road Extraction from Very High Resolution Images Using Weakly labeled OpenStreetMap Centerline

Fusing Taxi Trajectories and RS Images to Build Road Map via DCNN

Contact Info

Product

Resources

About