High-tech eyes check train 'fingerprints.' (CSX Corp. to use scanner to keep track of freight cars)

Sometime next month, a sophisticated scanner will begin helping a railroad company keep track of its freight cars. The system works by recognizing each car's distinctive features, says Robert Thibadeau, a computer scientist at Carnegie Mellon University in Pittsburgh.

Misplaced freight cars mean lost revenues, but cars do break down and get separate from their trains, so they don't always show up as scheduled. Railroad companies currently depend on employees to check car numbers as each train pulls into the yard. In addition, a small percentage of the 1.4 million freight cars in North America carry transponders that emit identifying radio signals.

But transponders break down, and ID numbers wear off as the cars undergo abuse from the elements. Moreover, updating the centralized database on car locations can take many hours. "People who are managing the flow of the trains . . . depend on the accuracy of the data," says Thibadeau.

Using machine-vision technology developed for industrial robots, Thibadeau has created a high-resolution image scanner that can identify a freight car, no matter how beat up. It notes physical traits as well as broken rivets, dents, rust, chipped paint or other types of damage that form a "fingerprint" unique to each car. The scanner's camera picks up details as small as a quarter of an inch, and its powerful computer anticipates degradation, taking those changes into account when identifying cars.

The system images a freight car as it passes into a railroad yard, picking up 9 megabits of data or more per car. If the fingerprint is not the one expected, the system checks through the fingerprints of all other cars expected in that train. If no match appears, then the system reads the identification number. If that fails, the program sorts through categories of cars and features to find the car's identity in the database. In theory, the computer can then instantly update a centralized database about the cars sighted.

The project's sponsor, CSX Corp., has set up one camera in Tampa, Fla., and may install others by November to assess the system's utility, says Percy F. Shadwell Jr. of CSX's Jacksonville, Fla., office. The company will evaluate how well the system holds up against foul weather and vandalism and whether it works fast enough to be practical.

"My personal opinion is that this technology as we're developing it may have a major impact on the whole transportation industry," says Shadwell. "The recognition technology we use could potentially be used to recognize anything." He also envision adapting the system for safety inspections.