Mobile phones as a guidepost

Image recognition software "FotoNav" facilitates orientation in foreign cities

Image recognition software "FotoNav" Author: Frank Luerweg / Uni Bonn
Read out

"Nothing but expensive bells and whistles" - camera phones do not deserve such a devastating judgment, say students from the University of Bonn and deliver the proof right away. Eight prospective computer scientists in the field of surveying have developed software that can identify the buildings depicted in digital images - largely independent of the viewing angle and lighting conditions.

The new technology could, for example, facilitate orientation in foreign cities: simply pick up a building, send it to a service number, and a few seconds later, the city map with the exact position appears on the mobile phone display. Although the Bonn project is not yet so far, interested parties can already test the software with their own images on their website .

The database still contains only facades from the Bonn Nußallee and the adjacent streets as well as from some of the city's most notable buildings. In the future, however, visitors should be able to send in their own pictures with address and description and thus be able to expand their photo inventory.

Image recognition software "FotoNav"

The operating principle of the image recognition software "FotoNav" is simple: it "distills" several hundred characteristic features from the digital data jungle - generally small image areas in which color or brightness change. If the software finds at least some of these features in the image database, it reports a hit.

"Differences in perspective or lighting can easily cope with the program, " explains Professor Wolfgang Förstner, Director of the Institute of Photogrammetry. In other words, even if the mobile phone owner picks up the building at an oblique angle, the computer can usually identify it. At just under 50 percent, the hit rate is now - not bad, considering that the project was created in a four-month software internship virtually "incidentally". "This is certainly much better, " says Förstner's doctoral candidate Marc Luxen; "We just did not have the time to work on the different sets of screws until everything runs smoothly." The development team now hopes to interest the industry in their process. display

Help from Canada

The original recognition algorithm does not come from Bonn, but from the Canadian researcher David G. Lowe. "The students should investigate the suitability of his basic idea for this particular application and optimize the method if necessary, " said Professor Förstner. Everyone paid attention to a different aspect, such as the speed of the process, the lowest possible memory overhead or the quality of the hit. "Unfortunately, it's like a Libra, " says internship participant Jens Külzer, "you pull on one side, and the other goes up. It was similar with us: what accelerates the evaluation usually comes at the expense of accuracy. But our goal was not the optimum, but to develop within the limited time a functional process. "

Among other things, K lzer has examined the economic perspective of camera-phone navigation. "At the moment, there is no reliable way to use mobile phones in inner cities, " he explains. "Even GPS positioning works only if the phone has a more or less clear view of the GPS satellites.

The method is intended to provide background information on interesting buildings

In narrow streets and tall buildings, this method therefore quickly reaches its limits due to shadows and reflections of the satellite signal on facades. "GPS receivers in the car, for example, therefore incorporate the driving speed in their cars Calculation with. This does not work for pedestrians - current cell phones are poorly aware of how quickly their owner is moving. "And also the position determination with the help of the active portable radio cell is quite inaccurate - in cities the Aufl sung amounts to approximately 50 to 100 meters." In the practice one could however for example use this Vorinformation to it, limit the search in the database to buildings in the corresponding mobile radio cell.

The process can do more than just determine the position of the mobile phone owner. K lzer: "It would also be conceivable to provide background information on certain touristically interesting buildings, which will then appear as text on the display - right in the appropriate language. The cell phone could even serve as a guide. "

(idw - Rheinische Friedrich-Wilhelms-University Bonn, 12.10.2004 - DLO)