Text, Lies, and Videotape
A picture may be worth a thousand words-but even a few words can help sort a thousand pictures. At Siemens Corporate Technology in Munich, Germany, engineers Andreas Hutter and Joerg Heuer have developed software that analyzes video sequences and automatically generates text descriptions of all the moving objects within them. Instead of classifying an object in a parking garage as a person-a difficult task-the software could indicate that a “triangle” (legs) with a “brown top” (hair) traced a certain trajectory. Armed with a text database, security personnel could then search for a car thief with brown hair walking along a certain path just by typing in a few keywords. The Siemens approach, which is compatible with digital-video data standards and could be commercially available in three to five years, is faster and more reliable than training a computer to search videos specifically for a walking person, says Hutter. And the computer doesn’t need to be reprogrammed if security also wants to look for something else, say a particular car: a search for a “red rectangle” with “circles” will do.