Machine vision solution developed to read Braille

Share this on social media:

In-situ, a Munich-based specialist for vision and sensor systems, has developed a vision solution for checking Braille text on medication packaging. The DotScan inspection system recognises Braille dots and evaluates their tactile quality with high precision.

Testing is non-contact using an industrial camera from IDS's USB uEye series. The monochrome CMOS camera captures 2D images. The height information is then calculated using a shape-from-shading algorithm. Lighting plays a key role in this process: to obtain accurate results, telecentric light sources are required, which illuminate the Braille embossed folding box from four directions at an angle from above. The telecentric light ensures that the 3D shape of a raised dot can be accurately derived from its shading pattern.

For the DotScan system, the system integrators at In-situ have developed custom lamps with a large field of illumination of 150 x 75mm. High power LEDs produce blue light by using a narrow band filter. ‘The advantage of blue light is that it has a short wavelength, so there is less reflection and that is better for analysing the raised dots,’ explains In-situ managing director Rainer Obergrußberger.

After embossing the Braille dots, the box is assembled and pasted. It is then folded flat and positioned on a drawer in the unit. The compact USB camera is positioned above the packaging. With a resolution of 1.3 Megapixels, the UI-1540SE-M model offers the ideal ratio of accuracy to data volume. The uEye captures 25fps at full resolution, which allows for fast inspection. The scene to be captured has a very high dynamic range, however. The shading has to be clearly discernible without overexposure in reflecting areas. Things are made even more difficult by the fact that the Braille dots are sometimes embossed on a black background, which is a major challenge in image acquisition. For this reason, the camera takes a series of images with different exposure times and superimposes them to determine the required dynamic range.

To accommodate the quick exposure time changes between the individual images, the developers at In-situ chose the software-triggered image acquisition mode. The camera was integrated using the proprietary uEye API programming interface.

The main challenge in developing the system was programming the algorithm. The shape-from-shading process was very slow at first and unsuitable for industrial use. After a long development phase, the developers succeeded in significantly improving the process: the image analysis, which initially took 10-20 seconds per image, can now compute up to 20 images per second. Obergrußberger said: ‘A lot of hard work and time went into the process. But we saw the potential of this method: Shape from shading offers the best price/performance ratio for this application.’

The DotScan system delivers an accuracy of +/-0.02mm depending on the surface properties of the folding box. The required calibration is carried out with special calibration standards.

In the field, the Braille inspection is performed by sampling. The system checks not only if the raised dots have the correct shape, but also if the information given in the Braille labelling meets the specifications.

Recent News

22 February 2021

Participants in the Embedded Camera API Exploratory Group will discuss requirements for new interoperability standards to accelerate market growth and reduce development costs

19 February 2021

On 4 March, panellists from Basler, MVTec, Sick, and Amazon Web Services will discuss developments in embedded vision during the Embedded World digital show. IMVE's Greg Blackman will moderate

12 February 2021

Cognex's revenue was $811m for the year, a 12 per cent year-on-year increase thanks to higher revenue from consumer electronics and e-commerce

10 February 2021

The firm's Lacera technology delivers greater than 90 per cent quantum efficiency and low noise architecture with up to 18-bit readout