Simultaneous Multi-Position and Multi-Part Vision Inspection with Low-Cost UVC Cameras

Abstract

Presence verification confirming the presence of a specific part in a designated location is a widely used vision-based inspection technique. However, conventional vision inspection systems are often equipped with expensive industrial cameras, which limits their wider use. This paper introduces a cost-effective presence verification system using low-cost UVC (USB Video Class) cameras. To tackle the inherent challenges arising from such cameras, we selected smartphone frames as a case study. We present a customized jig and data collection strategies to deal with the low-resolution and limited dynamic range issues inherent in low-cost cameras. Additionally, we perceive smartphone part inspection as a pseudo anomaly detection problem since acquiring non-attached parts samples or physically separating parts from already attached samples might not be feasible. In addition, considering the practical challenge of obtaining sufficient real-world samples and the ease of scaling for various parts and locations, we propose using conventional feature extraction and classification models. In our experiment, the proposed method achieved average performance of 99.96%, indicating its potential suitability for realworld industrial applications.