Statistical learning system for visual fault recognition.
The goal of this research and development project was to develop a visual inspection system with learning capabilities for recognizing small differences on products travelling on a production line.
For this we have developed a novel approach that uses statistical analysis of the images for dynamically separating the product from the background, and distribution analysis of pixels for filtering differences in the products.
The system comprises of one or more computers connected to one or more cameras mounted along the production lines. These different sites can be trained to filter faulty products in different stages of assembly. The system thus scales well for the task.
A pilot for the project was tested successfully by General Motors at Szentgotthárd, Hungary.