Defects Detection of Fluorescent Tube Based on Real-Time Image Texture Analysis

In this paper, an on-line machine vision system based on texture analysis is designed to detect defects of fluorescent tube. The fluorescent tube images can be captured by high-speed industrial cameras under certain lighting conditions. Key points location and shape approximation approaches are presented to get tube image region data timely; sliding window strategy is developed to extract local image features for the reason of discontinuity of global image intensity; image pre-processing and feature extraction algorithms are developed according to actual tube characteristics; variance and LBP texture description algorithms are presented to find defects in fluorescent tube images. The consume time of every tube image is less than 20 milliseconds and can meet the real-time detection requirements. Experiment results prove the effeteness and accuracy of our method; the detection device can be promoted to fluorescent tube production enterprises.