Entity

Time filter

Source Type


Pesek M.,University of Ljubljana | Strle G.,Znanstvenoraziskovalni center | Marolt M.,University of Ljubljana
Elektrotehniski Vestnik/Electrotechnical Review | Year: 2015

Several studies dealing with music recommendation and visualization base their approaches on datasets gathered with user surveys. However, the gathering procedure is seldom the focus of music research, even though the user interfaces and methodology are an important part of gathering the music data and evaluation of the music information retrieval algorithms. The paper presents the main elements of gathering the Moodo dataset that combines the demographic data, the users' mood and perception of emotions with the users' emotional and color responses to music. For this purpose, two novel user interfaces were developed, i.e. the MoodStripe and MoodGraph, which have several advantages over the existing classical models, both in terms of intuitiveness and functionality. The proposed interfaces are also applicable to other domains dealing with the user data.


Avbelj J.,FGG | Avbelj J.,TU Munich | Iwaszczuk D.,TU Munich | Stilla U.,TU Munich | Ostir K.,Znanstvenoraziskovalni center
Geodetski Vestnik | Year: 2012

The aim of this article is to investigate methods for the automatic extraction of the infrared (IR) textures for the roofs and facades of existing building models. We focus on the correction of the measured exterior orientation parameters of the IR camera mounted on a mobile platform. The developed method is based on point-to-point matching of the features extracted from IR images with a wire-frame building model. Firstly, the extraction of different feature types is studied on a sample IR image; Förstner and intersection points are chosen for a representation of the image features. Secondly, the three-dimensional (3D) building model is projected into each frame of the IR video sequence using orientation parameters; only coarse exterior orientation parameters are known. Then the automatic co-registration of a 3D building model projection into the image sequence with image features is carried out. The matching of a model and extracted features is applied iteratively, and exterior orientation parameters are adjusted with least square adjustment. The method is tested on a dataset of a dense urban area. Finally, an evaluation of the developed method is presented with five quality parameters, i.e. efficiency of the method, completeness and correctness of matching and extraction.


Obu J.,Alfred Wegener Institute for Polar and Marine Research | Podobnikar T.,Znanstvenoraziskovalni center
Geodetski Vestnik | Year: 2013

An algorithm of automated karst depression recognition uses a digital terrain model (DTM) and mainly applies the methods of a moving window with a kernel size of 3 × 3 cells using focal functions. It is divided into four parts: watershed calculation, depression delineation, higher level depression delineation and elimination of non-karst depressions. The essential part of the algorithm is the delineation of depression by the elevation of the lowest border cell of watershed. Depressions at higher levels are recognised by filling previously recognised depressions. The performance of algorithm was tested on test area in the Kras region (Slovenia) using DTMs with a spatial resolution of 12.5 m and 3 m. The results mainly depend on the DTM characteristics and quality, especially of their spatial resolution.


The paper describes the generation of a digital surface model (DSM) and orthoimages from panchromatic and multispectral Ikonos stereopairs. It assesses the suitability of the images for vegetation height mapping of a large area and the applicability of the results for various spatial analyses. The processing steps involved stereo bundle adjustment with various sets of ground control points, digital surface model extraction, orthoimage generation and evaluationof the results. Although both multispectral and panchromatic stereoimages were processed, the DSM was generated only for the panchromatic stereopair due to its higher resolution. For evaluation purposes it was compared to very accurate lidar elevation data. The analysis revealed an overall vertical difference between the models of 8.2 m, where only one third of the differences are below 3 m. The results were worse in steep areas with high vegetation and regions with shadows caused by hills or clouds. Better results can be obtained with previous manual or automatic editing of the automatically extracted model. On the other side, orthoimages that were also produced are very accurate - the evaluation showed results with horizontal RMSE errors below 1.5 pixels for both stereopairs when compared to aerial orthophotos.

Discover hidden collaborations