SIEMENS

Mail carriers and pharmacists have one thing in common: Both groups often possess cryptographic capabilities that enable them to decipher even the worst handwriting. Now, however, there are machines that can automatically recognize the most diverse types of writing — thanks to the development of amazing new learning processes.

The technology that makes this possible is known as optical character recognition, or OCR. “We’re the global market leader when it comes to address recognition,” Siemens product manager Peter Schindler says proudly. The capabilities here have nothing to do with the reading of machine-written texts, since any scanner can do that; the real accomplishment involves deciphering handwriting. Schindler estimates that OCR technology from Siemens is now being used at almost half of all mail sorting facilities around the world. The global market volume for these recognition systems currently stands at around a billion dollars, and Siemens’ Mobility Division, which manufactures the units, has a market share of 35 percent.

Siemens’ OCR developers are steadily improving the accuracy of their technology. “Our latest product line, ARTread, can decipher 90 to 95 percent of handwritten addresses,” says Matthias Schulte-Austum, the technical manager of the team that’s responsible for image preprocessing and object recognition at Siemens Mobility in Konstanz, Germany. But the system has to do more than just decipher messy handwriting. It also has to automatically identify all the relevant information on an envelope — things like changes of address, notes written on the side by the sender, and even the value of stamps. Postal automation systems also need to reliably recognize the sender’s instructions — for example, whether a letter should be returned to the sender if the addressee has moved.

The overall goal is to maximize the level of automation. “We want to automatically extract all the information relevant to the item in order to keep the amount of manual work as low as possible,” Schulte-Austum explains.

There’s still huge development potential for these systems, especially in Russia, India, China, and the Arab world. “We’ve developed algorithms that can read every kind of script, whether it’s Cyrillic letters or Chinese or Arabic characters,” says Ingolf Rauh, an expert at Siemens’ innovation center in Konstanz. “In fact, we recently won a competition for reading Arabic handwriting.” The challenge in the competition involved identifying the names of Tunisian towns without any mistakes.

The principles of optical character recognition are always based on the same rules. One method that has proved to be particularly efficient trains the systems to compare thousands of handwritten numbers or letters from various sources and then clearly classify them in the course of a learning process.

“We quickly recognized the great variety of potential applications for such a technique,” Rauh explains. “That’s why we decided to explore all the possibilities for using OCR technology, including those that involve completely new markets.”

Road Scanning. One such market is license plate reading for road pricing systems. For instance, Sicore systems from Siemens use cameras equipped with image-processing software to rapidly recognize license plate numbers as cars speed along streets and highways. Such systems are used in the UK, for example, where cities such as London have introduced a congestion charge. Cameras automatically register cars that enter a congestion zone and then check with a central database to make sure their drivers have registered with the fee collection system.

A further application involves using cameras that automatically record vehicles’ license plates in restricted-speed zones. Unlike radar guns, these cameras measure the average speed of vehicles over a long stretch of road. This enables the system to determine whether a driver has driven too fast through a tunnel, for example. “We’ve used our camera technology to develop a system called Safezone in cooperation with Siemens ITS in the UK,” says Stephan von der Nüll, who is responsible for developing new products and technologies at Siemens in Konstanz. “It’s the first system that makes this type of speed monitoring possible in inner cities.” Safezone is almost ready for market launch.

Tunnel Safety. An extension of the Safezone system is currently being evaluated within the framework of a project being carried out by Germany’s Ministry of Education and Research. Here, the goal is to automatically identify hazardous materials signs on trucks (see Pictures of the Future, Spring 2010, Danger Made Visible). The signs are orange and contain two numbers. The first number indicates the hazardous material’s classification, while the second identifies the hazardous substance itself.

“Automatic recognition of these signs will make tunnels and bridges safer,” says von der Nüll. Plans call for the system to automatically close a tunnel if, for example, a truck carrying hydrogen gets too close to another truck that is transporting oxygen. It will also be possible to quickly determine whether a truck with hazardous materials is about to entered a tunnel in which an accident has occurred.

It seems likely that enhanced OCR technology will be used practically everywhere in the future. Automatic recognition of food expiration dates and medications come to mind here, as does the identification of production and serial numbers on the printed circuit boards used in the automotive and electronics industries. Visually impaired people might also benefit, since OCR systems could read them their letters, books, or the food labels in supermarkets.