When it comes to vision systems, the most often presented is a 2D system, which is based on an industrial camera and special software.
2D-systems can be divided into 3 types:
With frontal illumination (reflected light mode). An image of the object itself is formed on the photosensor. The algorithms are based on video analytics methods.
Back-illuminated (shadow measurement method, transmitted light mode). An image of the shadow cast by the object is formed on the photosensor. In general, this method is significantly more accurate than the front-illuminated method, but is not always applicable due to limitations in the placement of the backlight. The algorithms are based on video analytics methods.
Based on neural networks with deep learning. They are used in tasks with complexly classifiable objects, when it is too difficult or impossible to use video analytics methods. For example, recognition of people, machines, or variations of complex objects (if the “pattern matching” method of video analytics is not applicable).