|Popis:||CMOS sensors that are used in vast majority of today’s consumer cameras, smartphones etc. use the rolling shutter (RS) mechanism to capture images. The key difference is that with the global shutter, the entire image is exposed to the light at once, whereas when using the RS the individual image rows (or columns) are captured at different times. When a RS camera moves while capturing the image, several types of distortion such as smear, skew or wobble appear. A perspective camera model is no longer valid in this case and that can be a problem when using methods assuming this model. Recently several algorithms for calibrating RS cameras have been proposed. These algorithms are usually evaluated on a very limited set of images (videos) without a ground truth. While there exist many benchmark datasets with ground truth camera calibrations and positions for perspective cameras, such datasets for RS cameras are still missing. The goal of this project is to create a benchmark dataset for RS cameras. The student/students will collect images/videos for different camera setups (a single RS camera, a stereo rig, two cameras with different rolling shutter directions or different frame rates, different types of camera movements (translations, rotations) during the image exposure….). Ground truth calibrations will be obtained using standard calibration methods, using a global shutter camera, and a controlled motion. For bachelor/master thesis the student will evaluate different state-of-the-art methods for RS calibration on the new proposed dataset. Based on the obtained results, the student will try to address challenging configurations (degenerate configurations) for different RS models.