Yeong-Sun Song

Road Infrastructure Data Acquisition Using a Vehicle‐Based Mobile Mapping System

This study presents the technology of a vehicle-based mobile mapping system to maintain an up- dated transportation database. The mobile mapping sys- tem that integrates the global positioning system (GPS), the inertial navigation system (INS), and digital cameras has been developed to collect data on position and at- tributes of road infrastructure. The vehicle-based mobile mapping system works by having the GPS and INS record the position and attitude data, and digital cameras take road images. The stereovision system can determine the position of objects that are visible on the image pair in the global coordinate system with GPS and INS data. As field data acquisition is a very expensive task, a mo- bile mapping system offers a greatly improved solution. In this study, we successfully created a road infrastruc- ture map with mobile mapping technology and proposed automatic algorithms for detecting and identifying road signs from road images. The proposed detection algo- rithm includes line and color region extraction processes and uses the Hopfield neural networks. The identification algorithm uses seven invariant moments and parameters that present geometric characteristics. With this combined method, we could successfully detect and identify road signs.

Radargrammetry for DEM generation using minimal control points

The erroneous satellite orbit causes significant distortions to the result of DEM (Digital Elevation Model) generated by radargrammetry. If accurate ephemeris data of satellite are available, satellite orbit parameters can be exactly determined without CPs (control points). If not, it must be refined by using the well defined CPs. For the refinement of orbit parameters, the acquisition of sufficient CPs is the most tedious and time consuming step and obstacle to near real time applications, especially for SAR (Synthetic Aperture Radar) imagery. This paper describes orbit parameters refinement method for radargrammetry using ephemeris data and only a single control point to resolve the CPs problem. The proposed method is applied to RADARSAT-1 SGF images of which ephemeris information is not accurate to directly use for accurate geometric modeling. The RMSE of the DEM generated by using a single CP (Control Point) is 43.86m, while that of the DEM generated by traditional method using 10 CPs is 45.79m in comparison with reference DEM that is generated from 1:5,000 digital topographic maps.

An efficient 3-d positioning method from satellite synthetic aperture radar images

This paper describes an efficient 3-D positioning method of SAR image only using ephemeris data and a single control point. The proposed method is applied to RADARSAT-1 SGF images of which ephemeris information is not accurate enough to directly use for the dynamic geometric modeling. The RMSE of the 3-D model by using a single CP (Control Point) is 43.86m, while that of traditional method using 10 CPs is 45.79m in comparison with reference 3-D model generated from 1:5,000 digital topographic maps of the study area. The test results show that even without the use of many CPs, it is possible to generate similar accuracy of 3D-model comparable to traditional method.

Accurate classification of water area with fusion of RADARSAT and SPOT satellite imagery

We fused RADARSAT images and SPOT panchromatic images by a wavelet transform in order to improve the accuracy of classification of water areas. Fused images not only maintained the characteristic of the SAR image (low pixel value) but also had improved boundary information. This leads to an accurate method to classify water areas.

Stereo matching strategy for 3-d urban modeling

This paper proposes an effective matching strategy to reconstruct 3-D urban models in densely built-up areas. Proposed scheme includes two main steps: feature-based image matching using building recognition technique and 3-D building reconstruction using the refined Rational Function Coefficients (RFCs). Especially, our approach is focused on improving the matching efficiency in complex urban scenes. For this purpose, we first performed automatic building recognition between stereo images, and then we endowed all points of building edges with identifiers using edge tracing method. Each identifier plays an important role in reducing search space for image matching within points of same building. A standard IKONOS stereo product was used to evaluate the proposed algorithms. It turned out that the proposed method could automatically determine the initial position and could dramatically reduce search space for point matching. Also, it was demonstrated that the updated RFCs could provide high-quality 3-D urban models.