Microview's solution uses an IBM Server X346, 3 Dell 5150n workstation computers, 3 Matrox Imaging Iris P1200 smart cameras fitted with a Pentax CCTV 12mm lens, 3 OEM light sources, 4 Santek MT500/1000 UPS devices, and one TP-Link 100M/16-port device. The image analysis is performed by the Coder Reader and String Reader modules from the Matrox Imaging Library. "The image processing software is extremely important to the application," explains Zhao. "Because the meters display both numerals and a bar code, we needed a solution that could handle the two different input types simultaneously." Zhao says they considered other machine vision vendors, but changed their minds when that meant purchasing two discrete vision systems, one for bar codes and one for character-reading. "With the Matrox Imaging Library we could read strings and the bar codes with a single machine vision system, and that allowed us to keep costs down," he notes.

Since the image capture process would be manual, Zhao and his team placed a high priority on simplicity and flexibility. The image capture software has two modes, one for the 1-phase meters and one for the 3-phase meters; the operator only has to reset the software type at the workstation to account for the different sizes. The meters themselves presented a more difficult situation. First, the thousands of different model types employ different fonts, which would make defining the font extremely cumbersome, if not impossible. Next, the meter's face features numbers that are not part of the dials (such as a serial number), which could interfere with the character-reading result. The plastic casing on the meter is often dirty, which can also interfere with the character-reading. Finally, the backgrounds of the dials are both black and white, which adds an extra step in specifying the read operation's settings. Finally, the white numeric dials with black numerals lie within a black rectangular frame which has the potential to confuse the read operation. The robust nature of MIL's String Reader module proved invaluable; defining/locating the string and their associated fonts is performed quickly, and the read operations are robust enough to read characters in poor or non-uniform lighting. In some cases, one of the meter's indicators may be between two values and cannot be 'read' by the String Reader; at such times, the operator must manually check the meter and record its value at the workstation.

"We could keep costs down by working with Matrox, " explains Zhao. "By developing with MIL we had all the image processing tools in one package and achieved very accurate results." Integrating the system with the Iris smart camera meant working with a single vendor, which prevented compatibility issues. Zhao's team programmed the application with MIL on the PC for image capture, pre-processing, barcode reading and character reading. Soon the application will be run on the Iris camera itself, since the Windows CE operating system facilitates programming on the camera directly. "We also received excellent technical support, which was very important to us," adds Zhao.

Presently, Microview has deployed a three-workstation system at Shanghai Electric Power Corporation, and there are plans to integrate an additional two workstations. They also hope to sell the system to Shanghai's other power utility. Considering features such as reading different input types simultaneously with a high level of accuracy, and the flexibility to develop the system as the customer's needs change, Microview's shouldn't have to try hard to find customers. But Zhao intends to get the most out of this technology and adapt it for other machine vision applications.