>>Vacancies for sandwich PhD students WHU-ITC

点击次数：

Vacancy announcement for two sandwich PhD positions of the Wuhan University (P.R. China) and theInternational Institute for Geo-Information Science and Earth Observation (the Netherlands)

The Wuhan University a key university under the direct administration of the Ministry of Education of the People's Republic of China. Both the State Key Laboratoryof Information Engineering in Surveying Mapping and Remote Sensing (LIESMARS) and the School of Urban Design are focusing on the education, R&D and social services through the whole cycle of geospatial information from sensor data collection to information system and its applications.

The International Institute for Geo-Information Science and Earth Observation (ITC) undertakes education, research and project services in earth observation and spatial information collection, analysis and management. ITC’s department of Earth Observation Science is a multi-disciplinary scientific team, specializing to maintain data collection and data analysis in earth observation, covering issues of mathematics, statistics, physics and computer science as far as these are relevant for earth observation.

The above two institutes invite qualified persons to apply for the positions of

PhD Student (f/m)

Dense 3D Description from Video
and/or
PhD Student (f/m)

Classification of multi-pulse high point density airborne laser scanner data over urban areas.

Main tasks

The focus of this research position is on the development, implementation and analysis of methods for acquisition of 3D geo-information from video or laser scanner data. More information is provided in the research descriptions on the next pages. Your research should result in a PhD thesis.

Requirements

The candidate graduated with excellent results, preferably on geo-informatics subject in geo-informatics, mathematics, physics, electrical engineering, or computer science. Good programming skills are a prerequisite. You are a team player with good expression skills, both orally and in writing. An IELTS score of 6.0 is required.

Planning

The position is planned to start at September 1st, 2006. You will first stay at ITC in the Netherlands for 18 months, including a qualifier period of 6 months. The following 12 months are spent at Wuhan University, followed by 6 months at ITC and 6 more months at Wuhan University. At the end you need to defend your PhD thesis at the University of Twente in the Netherlands.

We offer

We offer an inspiring and challenging international environment. For the period in the Netherlands ITC offers allowance for residential accommodation, meals, etc. (Euro 1084.50 per month), compulsory health insurance premiums, costs of residence permits and other essential permits, entitlement to accommodation at subsidized rates in the ITC International Hotel and a waiver for tuition fees. For the period in the P.R. China, Wuhan University offers the sameconditions as for enrolled graduate students for doctorates

Information

Additional information about this position can be obtained from Professor Zhu Qing (zhuq66@263.net) for the first position, Professor Zhan Qingming (qmzhan@126.com) for the second position or Professor George Vosselman (vosselman@itc.nl) for both positions. Interested candidates are invited to send their letter of application (in English), detailed CV and MSc course record before April 7, 2006 to one of the mentioned professors. You are invited to visit our homepages: http://www.whu.edu.cn/en/ and http://www.itc.nl.

 

Dense 3D Description from Video

Introduction

In the last years, there have been intensive research activities in the reconstruction of 3D point clouds from image sequences or image sets, especially in the field of computer vision [2]. Some commercial offerings are available, [1], [5]. However, the point clouds are not dense and do not allow a complete description of the 3D scene. We propose a research project which derives a dense 3D model from video data. The main objective is to fill the gaps in 3D point clouds by including oriented surface patches and edges. As an application, we consider the 3D modeling of streets. The video is taken by a moving car with a mounted monocular or stereo camera. The cameras are oriented sideways and capture the facades of buildings.

Outline

The outline of the proposed research is as follows:

• Image Point Matching and Reconstruction of relative orientation
  After image point extractionand initial matching, we suggest to use a RANSAC based approach for the reconstruction of the relative orientation, similar to [4].
• Computation of oriented surface patches
  Based on the extracted image points, the nextgoal is to compute the orientation of the 3D surface patches surrounding the image points. Oriented surface patches allow a richer description of the 3D scene than only image points. It is expected to utilize local region descriptors to derive the 3D orientation [3]. Local region descriptors are invariant to geometric and photometric image transformations and can be used to identify and relate image regions.
• Extraction of 3D edges
  For some of the extracted image points it is not possible to compute surface patches as they are corner points or edge points. In this case, one needs to detect these points and extend them along the edges.

Ensure consistency of patches, edges and points
It is important to ensure consistency along the extracted 3D surface patches, edges and points. This can be proven during the extraction process or as a final step.

References

[1] 2d3. http://www.2d3.com.

[2] R. I. Hartley and A. Zisserman. Multiple View Geometry in Computer Vision.Cambridge University Press, 2000.

[3] K. Mikolajczyk and C. Schmid. A performance evaluation of local descriptors. IEEE Transactions on Pattern Analysis and Machine Intelligence, 27(10):1615–1630, October 2005.

[4] D. Nister. An efficient solution to the five-point relative pose problem. IEEE Transactions on Pattern Analysis and Machine Intelligence, 26(6):756– 770, 2004.

[5] REALVIZ. http://www.realviz.com.

 

Classification of multi-pulse high point density airborne laser　scanner data over urban areas

Introduction

Over the past years the laser scanning technology improved considerably. Pulse rates for modern laser scanners are now in the range of 100-150 kHz, ten times the rates of scanners used in the nineties. The scanners now also are capable of storing multiple reflections of a single emitted pulse. The potential of these new features of laser scanners are hardly explored yet. It is expected that much better classification results can be obtained from this sensor data. It should also be possible to improve the outlining of detected objects. Not just because of the higher point densities, but also because all scanners are now combined with high resolution digital cameras. All these developments give good opportunities to automatically classify ground coverage in complex urban environments.

Outline

The main aspects of the proposed research are as follows:

• Extraction of features from laser scanner data.
  Planar features are the most dominant features in urban environments. Several algorithms have been proposed for the extraction of planes from point clouds. At the same time, connected component algorithms can also be useful to identify the smooth surfaces of the terrain or curved roof structures. Several algorithms are to be compared to obtain the best selection of features for a later classification step.

Segmented laser scanner data

• Extraction of potential building edges from multi-pulse laser scanner data.
  Edges of buildings as well as vegetation result in multiple reflections of a laser pulse. By selecting the pulses with multiple reflections and analysing the distribution of the points one should be able to extract locations of possible building edges.

Pulses with multiple reflections

• Classification of objects based on characteristics of segments and local point distributions
  Height texture measures and other characteristics of the extracted features should be used to make a classification into major categories (ground, buildings, vegetation). Depending on the point density more detailed classification categories (cars, people) could be included.
• Improvement of object outlining by fusion with edges extracted from digital imagery.
  The accuracy of edges extracted from (multi-pulse) laser scanner data is to be examined using photogrammetric reference data. After geo-referencing the laser scanner and optical data, edges extracted from the images should be matched with segment edges detected in the laser data and lead to more accurate boundary descriptions.