Screws in Plastic Casing

In this sample application, we present two ways to detect the presence of screws in the plastic casing of a portable cassette player.

The important point in this application is the essential equivalence of image objects and regions of interest in NeuroCheck. Many people fail to see this, because it is such a new and unusual way of thinking, but in our eyes, this really constitutes object-oriented image processing: not simply programming in a certain language, but using a consistent object model throughout each image processing function. Only in this manner you can reach the degree of interoperability and interactivity offered by NeuroCheck and create a system, in which each building block can work together with all the others to achieve hitherto unknown flexibility.

All screws present

One screw missing

One screw unfastened

The most obvious way for people familiar with NeuroCheck would be to search for the dark casing, which is kind of hard to miss, then go on searching for light objects within the casing, compute some features of these objects (like their size and area and shape parameters) and decide, whether two of them are the required screws. This is of course a perfectly viable solution, but the twofold object search and the evaluation of the objects take some time: 60 milliseconds on a Pentium 133 (not including image capturing, whose duration with a standard camera may vary within the video frame rate).

But it can be done within 16 milliseconds by using a vertical edge as position reference to place two regions of interest where the screws have to appear in the image. The presence of the screws can then easily be determined by the brightness inside the rectangle. It is even possible to detect a partly unfastened screw, because it sticks out and reflects too much light. Combine this with a DT3152 frame grabber and NeuroCheck's parallel image capturing capability and you get a real-time inspection!

Once again: whatever you can do with an object found in the image scene, you can also do with a manually defined region of interest. Of course, manual regions will behave differently in some respects, e.g. their position and size are fixed, but in all practical aspects they can be interchanged, and this is the cornerstone of the NeuroCheck's versatility.

We do not want to deny that there are other, more traditional ways of performing such an inspection. It is of course possible and at first sight perhaps even cheaper to build a mechanical solution, using two inductive calipers or something like that. But think of the future! A day will come, most probably not very far away, when the screws will be at a different position. You would be forced to a costly and time-consuming reconstruction of the mechanics, whereas moving the regions of interest in NeuroCheck is a matter of minutes. And it does not end there: the next day someone will want an inscription on the casing to be read, and how will you do that with a caliper?