World's Most Accurate Object Recognition Algorithm Developed in Maribor

Maribor, 9 January - Researchers at the Maribor Faculty of Electrical Engineering and Computer Science have developed an extremely accurate algorithm for object recognition. Using data from laser meters it allows the recognition of buildings, of the terrain and even of individual trees. Their algorithms are presently the most accurate in the world.

The algorithm was developed in the laboratory for geometrical modelling at the Maribor Faculty of Electrical Engineering and Computer Science. They are meant for the detailed processing of data from Lidar laser measuring devices, which are used in mapping and are placed onto planes or helicopters from where a 3D image of a surface is captured with a precision of a few centimetres.

The data collected is only a mass of unconnected points, researcher Domen Mongus explained. In order to create useful 3D surface models, these points need to be connected in a way that makes sense and different elements recognised - such as the ground, buildings, trees etc.

Such algorithms then enable the monitoring of large geographical areas. "We can determine individual objects, how many houses have been built, how the vegetation is developing, how quickly the trees grow. We can also follow major changes on the ground and carry out different kinds of simulations, such as flood simulations, or landslide monitoring," Mongus explained.

According to official results by the International Society for Photogrammetry and Remote Sensing (ISPRS), the team's algorithms are the most accurate in the world - Mongus said the accuracy exceeds 97%, which means they are on average wrong with less than 3% of the terrain points.

The accuracy of the algorithms even allows the recognition of individual tree crowns and of the trees' heights and if the measurements are repeated, the development of the tree can be observed.

"This means that we will at one point be able to look at each tree in Slovenia and determine its growth, its development through time and establish why trees grow faster in some areas and how much faster," Mongus moreover said about the use potential of the project.