Geometric Computer Vision
![Geometric Computer Vision](https://inf.elte.hu/media/42/af/6697f30da44190f08e7d973df7b9a1f91d2b83a1c8959bc218d99b48f67f/4-screenshot-eltecar-thumb.png?v202209270307)
Description of Activities
The main research field of the group is 3D machine perception. Principally, we process camera images, depth sensors, and LiDAR devices. The main active research topics are
- Calibration of different modalities, especially LiDAR devices and digital cameras.
- Stereo vision exploiting affine transformations.
- Visual odometry using vehicle-mounted sensors.
The application of affine transformations is the main focus of our research. This topic originated at ELTE.
The main application field is autonomous driving. At the university, a sensor-kit has been developed including cameras with narrow and wide field-of-view lenses, LiDAR sensor(s), IMU and RTK-GPS. The sensors are synchronized in time. A remotely controllable electric go-kart has been constructed here.
A programmable robotic arm is also developed for which the goal is the automatic unpacking of boxes; our research group developed a 2D LiDAR-based measuring software to detect the dimensions of the object that is gripped by the robot.
The group yearly organizes a competition for students, autonomous driving related tasks should be solved by the participants.
Our research work is supported by Robert Bosch GmbH.
Research Interests
Stereo and Multi-view 3D Computer Vision
- 3D Reconstruction
- Structure from Motion /Simultaneous Localization and Mapping
- Visual odometry
Multi-modal calibration
- Camera LiDAR
- Multi LiDARs
- 2D LiDARs
- RGB-D sensors
Research/service concepts/Methodology
We believe that theory and practice meet in modern research. Therefore, we address problems that appear in real-life applications. We prefer to use our own testing data, we do not only download those from the Internet. For this reason, we have developed (i) ELTECar: a car equipped with several sensors and (ii) ELTEKart: a remotely controllable electric go-kart.
Research Staff
- Levente Hajder (head), senior researcher
- Lajos Loczi, senior researcher
- Iván Eichhardt, senior researcher
- Bandó Kovács, engineer
- Tekla Tóth, PhD student
- Tamás Tófalvi, PhD student
- Máté Cserép, PhD student
Projects
- 3D reconstruction from rectified stereo images using affine transformations
- Machine vision for an unpacking robot
- LiDAR-camera calibration using spherical calibration objects
- Calibration of multi 2D LiDAR sensors
- Homography estimation for offside detection using the center circle
5 important publications in the field
- Tóth, T. and Hajder, L. (2023) ’A Minimal Solution for Image-Based Sphere Estimation’, International Journal of Computer Vision, 131(6), pp. 1428–1447.
- Hajder, L. and Baráth, D. (2020) ’Relative planar motion for vehicle-mounted cameras from a single affine correspondence’, 2020 IEEE International Conference on Robotics and Automation (ICRA), Paris, France, 2020, pp. 8651–8657.
- Baráth, D., Eichhardt, I. and Hajder, L. (2019) ’Optimal Multi-View Surface Normal Estimation Using Affine Correspondences’, IEEE Transactions on Image Processing, 28(7), pp. 3301–3311.
- Baráth, D. and Hajder, L. (2018) ’Efficient Recovery of Essential Matrix From Two Affine Correspondences’, IEEE Transactions on Image Processing, 27(11), pp. 5328–5337.
- Barath, D., Toth, T. and Hajder, L. (2017) ’A Minimal Solution for Two-View Focal-Length Estimation Using Two Affine Correspondences’, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 2017, pp. 2557–2565.
Infrastructure
ELTECar testing car and ELTEKart controllable go-kart
16-beam Velodyne VLP-16 LiDARs
HikVision 2 MPixel color cameras. With normal and fisheye optics
RTK compensation GPS
Intel Intellysense depth camera