Using Video Segmentation for Unsupervised Industrial Opitcal Production Control – Case Study for the SmartFactory@OST

This paper explores the application of video segmentation techniques for unsupervised industrial optical production control. It discusses the limitations of traditional manual inspection methods and the growing need for automated, efficient, and accurate quality control in modern manufacturing. The study focuses on unsupervised methods, particularly Meta’s Segment Anything Model 2 (SAM-2), for their adaptability to new product types without extensive labeled training data. One use case from the SmartFactory@OST is presented in detail: video segmentation for monitoring a cobot handling floorballs. The implementation demonstrates the potential of these techniques to enhance efficiency, reduce costs, and improve quality assurance in industrial settings. The paper concludes by highlighting the advantages of unsupervised segmentation methods and suggesting areas for future research and improvement.