A fast method for optical fiber defect detection and classification

Traditional manual fiber defect detection is inefficient and imprecise when the fiber moves fast, to solve the problem, a real-time optical fiber defect detection system based on machine vision is designed and developed. Detection system by three industrial cameras captures images of 0°, 120°, 240° angle in space which are transmitted to IPC to classify fiber defect. Fiber defects are defined to establish classification database and criterion. Common AdaBoost classifier is effective for this problem but wastes too much time, so an advanced AdaBoost cascade classifier based on morphological characteristics is designed. Detection results under industry condition show that the system meets the requirement of real-time detection and has high detecting accuracy of more than 99%.