DE9112775U1 - Arrangement for detecting irregularities in surfaces, such as recessed or raised characters - Google Patents

Description

Dr.-Ing. Hermann Tropf
Bluinenstr. 5

6837 St.Leon

Arrangement and ancestors for detecting irregularities in surfaces, such as recessed or raised characters.

The invention relates to an arrangement - ee for
Detection of irregularities in surfaces, such as recessed or raised written characters, in particular recessed or raised embossed/engraved/hammered characters or struck numbers.

For the identification or classification of workpieces, embossed characters are often used for reasons of durability
("Stroke numbers"). In the context of increasing automation, the task of automatically
controlled or read for production control purposes
Need to become.

The automatic recognition of embossed characters is subject to particular practical difficulties due to variable embossing characteristics, due to disturbances (scores, rust...) on the mostly
unprocessed surfaces, due to wear of the embossing tools, etc.

Automatic optical character recognition systems (they generally work with certain character models and features that either directly affect the geometry of the characters, such as character masks or
The more contrasting the characters are visually represented, i.e. the more they can be distinguished from the background, the more reliable the combinations of characters (e.g. with shape features, structural elements, directional elements or with other features derived from such geometric features) are.

In the case of metallic originals, illumination can generally only be carried out using incident light. Both brightfield and darkfield illumination are possible, see (/!/) .

A complicated process called a "retroreflector" was proposed for high-contrast imaging (/2/). The disadvantage for practical use is the complexity of the structure; the geometry of the workpieces and production equipment does not allow the use of such complex lighting equipment in most cases. In addition, maintenance and servicing considerations speak against their use. For industrial applications, the lighting equipment should be as simple as possible.

With simple bright field lighting, two main effects come into play:

1. Highlight method: This type of lighting is suitable for (almost) reflective surfaces that are not too disturbed. The mirror-like reflection is interrupted at the embossing points; the characters appear dark. The effect can be achieved by lighting and viewing at an acute angle or by an arrangement with a semi-transparent mirror.

2. Shadow formation: Shadow formation can be achieved simply by oblique lighting. This type of lighting has proven to be effective on matt surfaces (e.g. galvanized) and/or surfaces with strongly varying surface characteristics. The light is diffusely reflected on the surface; shadows are formed at the embossed points, which make the characters stand out darkly. With raised characters, a high-contrast pattern is created on the shadow side of the characters.

Both effects are used, for example, when checking embossed engine numbers, where the camera and lighting are at an acute angle to each other (/3/).

As experience shows, setting up the lighting for real-world punch mark recognition tasks is a very difficult matter (the optimal lighting direction depends, for example, on the main direction of any processing marks that may be present).

The task is to find an arrangement and a procedure that is less application-specific to install and yet still leads to reliable image evaluation.

During experiments with numbers on very rough surfaces (raw cast parts) that are particularly difficult to recognize with the naked eye, it was discovered by chance that, under very oblique bright field illumination, the vertical character structures stand out very clearly despite the very rough surface (shadows formed on the uneven areas), while the horizontal structures are barely visible - or vice versa.

The initially obvious idea of mathematically combining two separate images with different lighting leads to the following problem:

If you use two cameras that view the template from different angles, you get differently distorted images, and it takes a lot of computational effort to reconstruct the entire character in its complete form from two images (simple operations such as adding images after thresholding no longer work). The alternative of working with one camera but two lights leads, as experiments have shown, to partial images with less contrast.

The following idea led to a reliable yet elegantly simple overall solution:

To model the characters to be recognized, two separate sub-models are generated, each with a more vertical and a more horizontal structure. The image is captured one after the other using two separate camera/lighting pairs, with the viewing directions projected onto the object surface being arranged approximately or exactly perpendicular to one another, which essentially highlights the horizontal and vertical structures of the character with high contrast. The two sub-images are stored in separate image memories or separate areas of the same image memory. The character recognition algorithm works with separate features for each recording! The separate representation of the characters according to directional components leads to a representation that is generally not legible for humans in the difficult templates dealt with here; This may be the reason why the method has not yet been discovered by experts: in lighting experiments for image analysis tasks, one usually tries to first represent the objects in such a way that the resulting image is easy for people to interpret.

By separating the directional areas during image capture, direction-sensitive character features are displayed very clearly; horizontal and vertical structural elements do not interfere with each other; the detection of direction-dependent features and thus the image evaluation based on them becomes robust and reliable.

As an embodiment of the invention, the approach can be extended from 2 to n pairs of camera/lighting units with viewing directions rotated against each other, with the analysis of n partial images in order to be able to carry out a more differentiated directional analysis.

The approach described so far, which was found based on the problem of recognizing embossed characters, can be used in general to detect irregularities of complex geometry in surfaces.

Sources:

/1/: Ahlers, Rolf-Jürgen: Automated image processing. RKW-Verlag, Verlag TÜV Rheinland GmbH, Cologne, 1990, section 2.4.

/2/: "Making mistakes visible": Production, 1.12.1988 (No. 48), p. 16.

/3/: "Quality in the picture". Markt & Technik No. 25, June 16, 1989; p. 94; see illustration on page 95.

Claims (1)

Dr.-Ing. Hermann Tropf Blumenstr. 5 6837 St.Leon Expectations: Claim 1: Arrangement .and ancestors for the automatic detection of irregularities in surfaces, such as recessed or raised written characters, in particular recessed or raised embossed/engraved/hammered characters or punched numbers, characterized, da/3 &eegr;(&eegr;>=1) image recording devices (cameras) are used in incident light that are directed at the same surface, iUifeC and since/3 there is an illumination for each camera, but^two illuminations for the case n=l, Claim 2: Arrangement according to claim 1, characterized in da/3 the resulting individual images are kept in separate digital image memories or image storage areas.