Computer Vision Implementation in Scratch Inspection and Color Detection on The Car Roof Surface

Abstract

The automotive industry in Indonesia is one of the important pillars in the manufacturing sector. High speed productivity and high-level quality of their products are required for increasing the company value. The car roof inspection is still in low speed and inadequate quality control method for most of the car company’s manufacturing production line. This inspection process is highly dependent to human expertise and skill of the assembly operators.  This study purposes to increase the productivity in the inspection line by eliminate human error using computer vision. This research aims to find an automatic and accurate method for visual inspection of the color quality and scratch defect on car roof surfaces, using the data generated from the standardized database QR Code.  The proposed solution uses MobileNet-SSD method of computer vision in recognizing scratch defects, while color detection of the roof is attained through hue saturation method. The study employs data capture from a camera and compares the amount of object inspected in real time. The scratch inspection with 70 image samples and 4000 training steps results in a 3.33% error with color detection and inspection success rate of 80%.