Computer vision, mostly without AI
Industrial Problems Seminar
Elena Yudovina (CyberOptics)
Abstract
Most of the talks about computer vision these days will involve large neural networks, but people have been using computers to analyze images for decades before those came along. I'll talk about some of the image processing algorithms as applied to the systems used to inspect the electronics in your smartphone as well as in automotive applications; one of our customers had a manufacturing process that generated 5 faults per 25 million parts, and needed an inspection system that would contribute something useful in that environment. (For comparison, an error rate of 1 in 100 is typically considered a good result in image classification benchmarks.) I'll also mention some places where large neural networks may have something to contribute after all.