Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Fisheye camera calibration and depth estimation
The goal of this project is to develop a novel calibration model and method for fisheye cameras along with depth estimation algorithms based on this model.
Accurate depth estimation is crucial for reliable obstacle detection and avoidance in real-world robotic operations. To be cost-efficient, the robot’s environment should be observed with as few cameras as possible, advocating the usage of fisheye cameras due to their large field-of-view (FOV). Depth estimation based on standard mono- or stereo-cameras and the calibration thereof are well studied research topics. The calibration and depth estimation becomes considerably more challenging for cameras with a FOV around 180°, due to the much stronger lens distortion. However, the successful modeling and calibration of such a camera can have a huge impact: with little hardware involved robots may survey their complete environment - an impressive example is e.g. the Skydio2 where a few cameras provide full surround view.
The goal of this project is to develop a novel distortion and/or calibration model and find an efficient method to integrate it into an existing framework for depth estimation. Ultimately, we want to achieve accurate depth estimation despite strong lens distortion. To this purpose, the student may investigate cutting-edge camera models and calibration methods, assess what distortion models can leverage from recent developments in Machine Learning / Deep Learning, as well as depth estimation algorithms. She/he will be invited to contribute to a common framework and will have the chance to work with close-to-market robot prototypes. Finally, by the end of the project, the student will have developed considerable knowledge in the challenging topics of depth estimation, camera modeling and calibration. The hands-on experience with a mobile robotic platform will further augment this experience with a deep understanding of the software-hardware interplay.
Accurate depth estimation is crucial for reliable obstacle detection and avoidance in real-world robotic operations. To be cost-efficient, the robot’s environment should be observed with as few cameras as possible, advocating the usage of fisheye cameras due to their large field-of-view (FOV). Depth estimation based on standard mono- or stereo-cameras and the calibration thereof are well studied research topics. The calibration and depth estimation becomes considerably more challenging for cameras with a FOV around 180°, due to the much stronger lens distortion. However, the successful modeling and calibration of such a camera can have a huge impact: with little hardware involved robots may survey their complete environment - an impressive example is e.g. the Skydio2 where a few cameras provide full surround view.
The goal of this project is to develop a novel distortion and/or calibration model and find an efficient method to integrate it into an existing framework for depth estimation. Ultimately, we want to achieve accurate depth estimation despite strong lens distortion. To this purpose, the student may investigate cutting-edge camera models and calibration methods, assess what distortion models can leverage from recent developments in Machine Learning / Deep Learning, as well as depth estimation algorithms. She/he will be invited to contribute to a common framework and will have the chance to work with close-to-market robot prototypes. Finally, by the end of the project, the student will have developed considerable knowledge in the challenging topics of depth estimation, camera modeling and calibration. The hands-on experience with a mobile robotic platform will further augment this experience with a deep understanding of the software-hardware interplay.
- Make yourself familiar with our robotic calibration and perception framework as well as the current state-of-the-art.
- Build upon the state of the art by developing your own ideas and your supervisor's input.
- Design, test, and deploy a calibration method for fisheye cameras.
- Design, test, and deploy a real-time capable solution for depth-estimation using your developed model.
- Design and conduct experiments with a mobile robot to evaluate the selected approach.
- Make yourself familiar with our robotic calibration and perception framework as well as the current state-of-the-art. - Build upon the state of the art by developing your own ideas and your supervisor's input. - Design, test, and deploy a calibration method for fisheye cameras. - Design, test, and deploy a real-time capable solution for depth-estimation using your developed model. - Design and conduct experiments with a mobile robot to evaluate the selected approach.
- Strong self-motivation and curiosity for solving challenging robotic problems
- Previous experience in computer vision (calibration and/or depth estimation)
- Good C++ and Python programming skills, knowledge in ML/DL is a plus.
- Experience with Linux, ROS, and typical development tools such as git 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 problems - Previous experience in computer vision (calibration and/or depth estimation) - Good C++ and Python programming skills, knowledge in ML/DL is a plus. - Experience with Linux, ROS, and typical development tools such as git 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 Yannick Huber (yannick.huber@sevensense.ch), Thomas Eppenberger (thomas.eppenberger@sevensense.ch), and Renaud Dubé (renaud.dube@sevensense.ch).
If you are interested, please send your transcripts and CV to Yannick Huber (yannick.huber@sevensense.ch), Thomas Eppenberger (thomas.eppenberger@sevensense.ch), and Renaud Dubé (renaud.dube@sevensense.ch).