SightLine: A Low-Cost AI Vision System for Real-Time Defect Detection in Manufacturing

Automated inspection of visual quality is heavily reliant on machine-vision technology used in industrial plants, which can be too expensive for small and medium-sized businesses. The technology detailed here—SightLine, an inexpensive AI-based vision system capable of recognizing visible product defects through the use of common web camera and smartphone devices—is designed to automate visual quality inspection processes. The system takes images of produced products and processes these images using a unique predetermined algorithm to classify a specific product as good or defective with the possibility of locating the defect in a product with the use of a bounding box. In a trial using 4,900 labeled images from five categories of defects, the selected method produced a mean average precision at intersection-over-union 0.5 (mAP@0.5) of 0.912, with a per-frame inference delay of 24 ms using a 15 W device. We examine the compromises between performance, time, and the price of hardware to determine the circumstances in which using low-cost machine vision can be realistically adopted rather than the traditional method of inspection. All figures of merit are based on prototypes and will have to be verified later on the real production line of the factory.