Automatic gathering of labeled images using a web crawler to facilitate automotive machine learning

Abstract

Data preparation for machine learning is time consuming. We built an efficient machine-learning system and automated data preparation. Using a Web crawler, multiple keywords were used to download images to facilitate machine learning of automotive applications by local computers. A verification module then filtered the target images. Here, we explored weather while driving using only images containing vehicles. We found it simple to download images using a Web crawler without manually preparing the experimental learning data. We verified the utility of our approach using a computer vision algorithm. It is possible to eliminate most passive machine learning work when creating huge amounts of diverse data. Our method automatically generates a machine learning dataset for exploring weather features by capturing forward images taken from vehicles.