With the latest member of its VISOR® family, SensoPart is making vision applications even easier. Thanks to the artificial intelligence bundled in the «Classification (AI)» detector, the VISOR® Object AI independently learns distinguishing characteristic features based on a few images of the object to be detected. Even strong process and product variations such as fluctuations between batches, contamination, reflections, changing shapes or varying 3D orientation can be taught with just a few mouse clicks. It’s then able to reliably recognize the objects appearing in front of the lens and assign them to different classes.
For presence checks, components can be rated as «good» or «bad» or divided into 200 classes – for example, to ensure that the right parts for the respective product are always supplied and processed for product variants. Once a classification has been taught, it works reliably and robustly, without the user having to worry about suitable detection rules and parameters, as is the case with classic, rule-based image processing (e.g. using pattern comparison, contour or contrast recognition). Because the VISOR® Object AI is capable of learning, it typically only needs around five sample images per object class to sufficiently achieve a stable detection process. The AI algorithm is implemented in the sensor itself and therefore does not require any network or cloud connections.
Solves problems that would otherwise be difficult to solve
The application possibilities of SensoPart’s new AI vision sensor are just as diverse as its built-in classification competence. In automobile production, it can differentiate between component variants and determine whether the appropriate variant is available for a specific vehicle equipment. When flexible, shape-changing objects such as spiral springs or plastic bags are fed in, it detects wrong parts or incorrect positions.
Compared to classic detectors, the AI vision sensor can solve such tasks with significantly reduced setup effort and increased process stability. The user saves time because he does not have to create a logical link between several detectors.