Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Line Clustering and Description for Place Recognition
The goal of this project is to develop a place recognition pipeline using RGB-D or stereo cameras, by extracting line clusters.
Keywords: Computer Vision, RGB-D, Stereo, Line Clustering, Line Description, 2D, 3D, Place Recognition, Object Detection, Machine Learning
Most of the traditional SLAM or object recognition techniques rely on detecting salient point features in images. The main requirement for these points is that they are distinctive enough to uniquely identify an object/place, and repeatable enough to find the same points (in the same location) in different images. This approach has proven to be successful, however, some challenges such as textureless regions, large viewpoint changes, occlusions, changes in appearance due to weather, illumination, season, etc., still remain.
Since lines are more robust to environmental changes, are effective in low-textured scenes, and provide complementary information to point features, they have potential to tackle some of these challenges and provide a more robust system. Furthermore, by extending the line detection and description from 2D images to 3D point clouds, more information can be obtained and more distinctive features can be computed. Finally, instead of relying on individual lines, clusters of lines might provide much more coherent and unique features that could be used for either place recognition or object detection.
In this project, the goal is to explore and implement different strategies for clustering lines in 3D and describing such clusters. Afterwards, an indoor and urban outdoor dataset, will be used to test the place recognition pipeline. Finally, if the results are promising and time permits, the working pipeline can be plugged into a SLAM backend and evaluated. Finally, depending on the project outcome, the student will be invited to publish his/her work.
Most of the traditional SLAM or object recognition techniques rely on detecting salient point features in images. The main requirement for these points is that they are distinctive enough to uniquely identify an object/place, and repeatable enough to find the same points (in the same location) in different images. This approach has proven to be successful, however, some challenges such as textureless regions, large viewpoint changes, occlusions, changes in appearance due to weather, illumination, season, etc., still remain.
Since lines are more robust to environmental changes, are effective in low-textured scenes, and provide complementary information to point features, they have potential to tackle some of these challenges and provide a more robust system. Furthermore, by extending the line detection and description from 2D images to 3D point clouds, more information can be obtained and more distinctive features can be computed. Finally, instead of relying on individual lines, clusters of lines might provide much more coherent and unique features that could be used for either place recognition or object detection.
In this project, the goal is to explore and implement different strategies for clustering lines in 3D and describing such clusters. Afterwards, an indoor and urban outdoor dataset, will be used to test the place recognition pipeline. Finally, if the results are promising and time permits, the working pipeline can be plugged into a SLAM backend and evaluated. Finally, depending on the project outcome, the student will be invited to publish his/her work.
- Make yourself familiar with current line detection algorithms.
- Extend or implement a new line detection algorithm for stereo and RGB-D cameras.
- Perform a literature review on (semantic) clustering.
- Select and implement the most promising strategy for performing line clustering.
- Implement a technique to describe the clusters.
- Design and conduct experiments to evaluate the selected approach.
- Make yourself familiar with current line detection algorithms. - Extend or implement a new line detection algorithm for stereo and RGB-D cameras. - Perform a literature review on (semantic) clustering. - Select and implement the most promising strategy for performing line clustering. - Implement a technique to describe the clusters. - Design and conduct experiments to evaluate the selected approach.
- Strong self-motivation and curiosity for solving challenging robotic perception problems.
- Excellent programming skills (i.e. having written several thousand lines of code) ideally in C++ and the ability to work on large code bases.
- In-depth knowledge in at least two of the three following areas: Machine Learning, Optimization, and Computer Vision.
- Experience with Linux, ROS, and typical development tools such as GIT or Jenkins are advantageous.
- A very good academic record is desirable but may be compensated by expert knowledge in the areas mentioned above.
- Strong self-motivation and curiosity for solving challenging robotic perception problems. - Excellent programming skills (i.e. having written several thousand lines of code) ideally in C++ and the ability to work on large code bases. - In-depth knowledge in at least two of the three following areas: Machine Learning, Optimization, and Computer Vision. - Experience with Linux, ROS, and typical development tools such as GIT or Jenkins are advantageous. - A very good academic record is desirable but may be compensated by expert knowledge in the areas mentioned above.
If you are interested, please send your transcripts and CV to Fadri Furrer (fadri.furrer@mavt.ethz.ch), Florian Tschopp ( florian.tschopp@mavt.ethz.ch), and Tonci Novkovic (tonci.novkovic@mavt.ethz.ch).
If you are interested, please send your transcripts and CV to Fadri Furrer (fadri.furrer@mavt.ethz.ch), Florian Tschopp ( florian.tschopp@mavt.ethz.ch), and Tonci Novkovic (tonci.novkovic@mavt.ethz.ch).