CNN을 이용한 스티어링 휠 이종 불량 검출에 관한 연구

Abstract

Steering wheel that we handle the most driving cars has many switches not only with steering function but with controlling a number of other functions within the cars. Therefore, Steering wheel is equipped with completely different switches even for the same model according to consumers’ choices for options. Steering wheel manufacturers produce diverse steering wheels following consumers’ needs and then supply those to car manufacturers but very often the human error happens unintentionally, when option “A” needs to be assembled, option “B” comes out as a result, due to various options on the list. Vision Test Machine is used to avoid this human error, but It works simply in the way of comparing images and therefore miscalculations caused by micro conditions occur which are obvious demerits not to mention being inefficiency. In this paper, I present one of the deep learning models that shows outstanding performance in classifying images and assorts car steering wheels using Convolution Neural Network. Steering wheel images collected from 2016 to 2019, made by Vision Test Machine in Steering wheel manufacturer. All collected images are defined as 5 categories. One category of them is used in this paper. Not the whole picture of the image, only 3parts of an image which have an effect on classifying steering wheel are extracted and used. Convolution Neural Network uses a model based Inception module and a model based ResNet. Google’s deep learning open source framework, Tensorflow, is used for the experiment and OpenCV and Scikit-Learn are used for preprocessing images and processing data. Steering wheel categorization-model proposed in the paper can be applied in Vision Test Machine System or in actual work adopting additional data processing steps