Inspection System for Vehicle Headlight Defects Based on Convolutional Neural Network

This paper proposes a method to detect the defects in region of interest (ROI) based on convolutional neural network (CNN) after alignment (position and rotation calibration) manufacturer’s headlights determine whether vehicle are defective. The results were compared with an existing for distinguishing among previously proposed methods. One hundred original headlight images acquired each two types purpose this experiment, 20,000 high quality defective obtained by applying position transformation images. It was found that demonstrated performance improvement more than 0.1569 (15.69% average) as method.