Lukás Cerman

Learn More
It is suggested how a Markov random field can be used for object tracking with context information. The tracking is formulated as a two layer process. In the first phase, the image is represented by a set of feature points which are tracked by a standard tracker. In the second phase, the proposed semi-supervised learning and labeling algorithm is used to(More)
The limited dynamic range of a camera may be extended by composing differently exposed images of the same scene. The nonlinear camera has to be calibrated radiometrically first. We show that the calibration process can be difficult for real cameras. The improvement can be achieved if linear 12-bit RAW images are used as offered by modern mid-class and(More)
To build a full 3D model of a physical object, multiple partially overlapping parts of an object model need to be merged. Since modern range finder devices provide color information in addition to the depth information, the color can be used to help the registration process as well as to qualitatively improve the reconstructed model by incorporating(More)
Tracked objects rarely move alone. They are often temporarily accompanied by other objects undergoing similar motion. We propose a novel tracking algorithm called Sputnik Tracker. It is capable of identifying which image regions move coherently with the tracked object. This information is used to stabilize tracking in the presence of occlusions or(More)
The limited dynamic range of a camera may be extended by composing differently exposed images of the same scene. The nonlinear camera has to be calibrated radiometrically first. We present several state of the art calibration methods and show that the calibration process can be difficult for real cameras. The problems with calibration can be overcomed if(More)
A histogram-like model is suggested for the representation of multi-dimensional distributions such as RGB colors of subjects in tracking and segmentation tasks. Unlike the normal 3-D histogram it can be estimated from the limited amount of training data without the need to reduce the precision of the measured data. The proposed hierarchical histogram model(More)
  • 1