OCR System Puts DOT on Course

Systems integrator ImageSource provides a fix for Washington state's accident-reporting data troubles.

Brian Limotti, project manager for the state of Washingtons Department of Transportation, found his agency under the gun in 2001. In the 1990s, traffic safety officials in Washington had embarked on an ambitious program to modernize the analysis of accident-report data. They had ended up pretty far off course.

The State Patrol (Washingtons name for its state police) then spent four years trying to implement OCR (optical character recognition) technology so it could automatically scan troopers written accident reports into a database.

Alas, OCR wasnt ready for prime time, and by 2000, the project was considered a boondoggle. Worse, state accident data fell years out of date, risking the loss of precious federal highway funds.

"If a state is habitually behind schedule, the federal government tends not to view it very favorably," Limotti said. "Washington had been living on borrowed time here."

In 2001, Limottis agency was charged with fixing the mess. The legislature appropriated $1 million for the project; in exchange, it wanted a system that would make all accident reports available to the public within two weeks of a trooper submitting them and data from those reports compiled into DOT databases within three months.

It took three years, but along the way the DOT discovered a vital truth: Sometimes a little manual entry can save a lot of automation cost.

To tackle the problem, the DOT turned to ImageSource Inc., a systems integrator based in Olympia, not far from Limotti and other state officials. The two sides conferred for six weeks at the end of 2002, studying DOT workflow patterns and reviewing the projects past wrong turns. "On a frustration scale of 1 to 10," said Limotti, "the previous OCR system was a 20."

Shadrach White, chief technology officer at ImageSource, said the biggest mistake made by state officials was to assume that OCR technology would abolish the need for manual entry of report information. In reality, OCR failed to record data on the handwritten reports (sometimes missing whole pages) so often that the State Patrol had to hire more staff to make corrections.

"Clients in those days thought you took a big stack of paper, threw it in the scanner and the software does the work for you," White said. "Even today that wouldnt work."

Washington generates approximately 150,000 accident reports annually, Limotti said. State troopers, county police and city police all use the same basic form, which has 120 data fields for victim names, car type, road location, weather and more.

With that many data points to track, ImageSource turned to Kofax Image Products Inc., a maker of imaging and data-capture applications based in Irvine, Calif., and carefully mapped out what data should be entered manually to avoid making mistakes from the beginning. The thinking, White said, was simply to cut OCR and its risky consequences out of the data-recording process so workflow software could route reports with less delay.

"That was the linchpin. As soon as we got them away from correcting OCR errors and directly keying in information, productivity skyrocketed," White said.

Next Page: The project begins anew.