• Používáte starou verzi internetového prohlížeče. Doporučujeme aktualizaci na nejnovější verzi. Některé funkce portálu nemusí být podporovány.

• Pravděpodobně máte vypnutý JavaScript. Některé funkce portálu nebudou funkční.

“A digital photo contains approximate GPS coordinates of the position, where the photo has been taken. We need to specify this position and find a view direction of the camera. We can do that with digital landscape models, that provide us with altitude data of the landscape. Such 3D models of Earth are freely available, we can easily determine the location of forests, mountains or glaciers. We use silhouettes to compare the picture with the 3D model to determine the precise location and view direction of the camera,” Martin Cadik explained the principle of visual geo-localization of photos and videos. His team is the only one in the Czech Republic focusing on landscape pictures and one of very few researchers in the field in Europe.

Geo-localization of pictures also has uses in the security industry. A precise localization of a picture can help emergency services localising a missing person. Experts can determine a precise location of a picture from the features of surrounding terrain. Such information can also be used in forensic analysis in determining whether a picture contains what its author claims and whether it has been taken on the claimed location (such as peak of Mt Everest).

Members of Adobe Research, including Mike Lukac, are collaborating with Martin Cadik´s team on the research. Cadik noted that as part of the work together, his doctoral students would gain a unique opportunity at an internship at one of the world’s best-known software companies. Doctoral student Jan Brejcha from Cadik´s research team has completed such internship recently.

Apart from forensic experts, ordinary users can benefit from geo-localization as well – for example to edit their holiday pictures. Computational photography offers a wide range of uses in editing existing pictures. “The user can simulate a different camera and refocus an existing picture. Information on the distance of a mountain peak can be used to artificially refocus the picture in a way our equipment would never permit. We can see, how a picture would look like if we would take it with a large telescope,” said Cadik, a former Max Planck Institute post-doc, presenting a possible application of his research.

“A picture dimmed with mist can be easily and automatically brightened and skies cleared. On the other hand, a mist can be easily added for an artistic impression. Nowadays, this is a difficult task to be done manually, we know how to achieve this automatically,” said Cadik, who claims that computational photography brings a new approach to editing pictures. It is possible to change the lighting of a picture, add artificial shadows made by the terrain features and edit the picture to look as if it was taken during sunrise or sunset. “Nowadays, it is increasingly difficult to recognize an edited, artificially altered picture. However, it is possible. That is the focus of our work in forensic picture analysis,” added IT researcher from Brno.

A connection of information from a picture and 3D terrain models help users gain more data on surrounding landscape, such as names of nearby mountains, distances and altitudes of their peaks. Geo-localization of pictures could also be utilised by Swiss Alpine Rescue service that uses extensive camera network to monitor snow cover. An extension of data from digital 3D models could provide rescuers valuable additional information on the surrounding landscape.

Information on landscape is also helpful with navigating self-driving cars and drones. Martin Cadik´s team focuses on geo-localization of landscape pictures, however they can also analyse photos and videos taken in cities as well. In such cases, they compare a picture with an existing database, such as Google Street View. With landscape models, the researchers use digital 3D models enhanced by data from a public geographical database Open Street Map. Geo-localization is also relevant in the design of autonomous space rovers used in Mars exploration, that also use surrounding landscape features for navigation.