Förslaget inkom 2008-05-30
Vision-based Recognition, Categorization and Semantic Labeling of Places for Mobile Robots
OBS! ANSÖKNINGSTIDEN FÖR DETTA EXJOBB HAR LÖPT UT.
One of the fundamental competences for a mobile robot is the ability to localize itself, that is find its position in the world. Most existing methods rely on range sensor information for this task, e.g., from a laser scanner. However, humans are a living proof that the same task can be done using visual information. In an ongoing research project an algorithm for visual place recognition based on state-of-the-art machine learning techniques has been developed. So far the evaluation of this method has been limited to one floor of the CVAP lab at KTH, and the main challenge has been achieving robustness to varying illumination conditions, and to the natural variations that occurs in time as a room is used. An equally important, but almost unexplored issue, is the capability to generalize across different environments: if the system is able to recognize my office, would it be able to recognize your office? And what about any office? The objective of this thesis project is to extend the evaluation of the method to the generalized task of place categorization (recognizing AN office, not THE office), and investigate the performance on data collected from four different labs in Europe. Can the same methods and settings be used for both recognition and categorization tasks in all locations? Additionally, applicability of state-of-the-art techniques in visual recognition based on the idea that the visual world can be decomposed into a set of reoccurring structures (visual words) will be explored in this new scenario.
The first part of the thesis project would be carried out at KTH and would involve reading up on the literature as well as acquiring and preparing the data for the analysis using mobile robot platforms. After that the student would be given the opportunity to visit IDIAP in Switzerland for a few months to work on the analysis. The work would then be wrapped up in Stockholm where the report would be written.
Requirements: Basic knowledge in pattern recognition and digital image processing or signal processing. Experience with Linux-based computing environments. Programming skills (C/C++, Matlab (a plus), scripting languages).
Applications should be made as soon as possible. Applicants are invited to email a CV, study report, and a short application summarizing the personal interests and skills in
- pattern recognition / machine learning
- image processing / computer vision
- programming, computing environments, and operating systems
Additionally, please specify the earliest date when the work could be started.