Projects

Reading best-before dates on yoghurt pots is a challenge for classic image processing systems, as the writing on a curved background is often distorted and varies in size and contrast. With the help of deep learning, letters and numbers can be read reliably in different light and background conditions.

Classic image processing quickly reaches its limits, especially in the area of ​​surface inspection. With the help of the Neuralzye® software from senswork, an AI inspection of transparent strip material has now been implemented, with which the detection of bubbles and inclusions can be detected.

A 360 ° panorama endoscope is used to examine the inside of an aluminum body for processing marks and surface defects such as cavities and pores. Our system checks cylindrical internal surfaces in several height sections using deep learning tools for irregularities in the surface.

Chocolate is difficult to describe completely with a rule-based algorithm, since no two chocolates are identical. The individual chocolates have a high natural range of variation.

The individual elements of a first aid kit must be identified through a transparent packaging. Since the packaging reflects light, the individual components of the packaging are difficult to recognize. ViDi Check inspects the complete first aid kit and determines whether all trained features are present.

Varying the font and character size on a non-homogeneous background is very difficult for conventional OCR. With a robust training set, ViDi Read can accurately read letters in a variety of lighting and background conditions.

Varying opacity and bubbles in the liquid create a very inconsistent background in the bottle. The imprinted date and batch codes are difficult to recognize for ordinary image processing. ViDi Read can read letters accurately under a wide range of light and background conditions.

Different liquids create inconsistent backgrounds under the transparent bottle. Plastic seams and difficult reflection angles pose another challenge. ViDi Read can accurately read letters under diverse light and background conditions.

An inhomogeneous topology, different shades and the changing shape of the tortilla make the detection of cracks or wrinkles extremely difficult. ViDi Detect can distinguish between tolerable deviations and defects by training on a reference set of approximately 100 different tortilla samples.

A curciut board for LEDs is very complex and can contain a large number of errors. This makes it difficult to describe all the required features of a good part with rule-based image processing. ViDi Detect uses an example-based approach to model the full range of required circuit board features.

The transparency of glass vials makes it difficult to distinguish breakage from dirt along the front or back. Rule-based image processing reaches its limits here. ViDi Detect can accurately distinguish the variation of natural parts from anomalies by performing a reference set of good parts under varying light conditions and amounts of dirt.

Badly positioned, incorrect or bent components on a populated circuit board are difficult to detect with conventional image processing. Using Deep learning, classification and defect detection is much more reliable, even with a low even with a small amount of training data.

Before spark plugs are packed, they are inspected for completeness per layer. The space situation only permits the use of wide-angle optics, which leads to a strong distortion of the perspective. In this case, classic pattern searches can already reach their recognition limits. Using Deep Learning, however, the the inspection works very reliably.

Defects in air filters must be detected on a complex structured surface. Fluctuations in the light reflection make the use of rule-based image processing impossible. ViDi Detect accurately detects errors in the structured area - despite a high natural variance of the air filters.

Stamped metal at different angles of reflection presents a challenge for rule-based image processing: No two features look exactly alike,especially on curved surfaces. ViDi Read can accurately read letters under a wide range of lighting conditions.