OCR (Character Reading)

Industrial character reading poses different challenges from document OCR: characters may be embossed in rubber, engraved in metal, laser-marked on plastic or printed with ink, and the contrast may be low and the surface curved and glossy. The DOT code on a tire sidewall is one of the hardest cases: raised black characters on black rubber.

In our tire application the camera images the area of the sidewall carrying the DOT code; OcrVision processes the image to bring out the raised characters, reads the code and sends the result to the PLC. If the code cannot be read, the system raises an alarm. The operator screen shows the code read with its confidence, the cycle time split into camera, reading and PLC times, and the OK and NOK counts for the shift; in the examples on this page the total cycle is about one second.

OcrVision is based on deep learning: we train the reading model for each customer with images collected on site and shape it to the product, character style and surface conditions of that line. As new data comes in from the field, the model can be retrained and improved.

When the position of the code is unknown, the whole sidewall is imaged, the circular surface is unwrapped into a flat strip, and the DOT code is located and read on that strip, so it does not matter at which angle the tire arrives at the station. The same approach applies to serial numbers, part codes, dates and lot information on metal and plastic parts.