부산대학교 도서관

Recently, technologies related to autonomous vehicles have been developing rapidly. In line with the advancement in autonomous driving, fully autonomous vehicles that can be driven without human intervention will soon be commercialized. When fully autonomous vehicles appear, various in-vehicle services will change along with automobile culture. Autonomous vehicles may offer services for enjoying multimedia content and entertainment in the vehicle. In this paper, therefore, we propose a method for recommending multimedia to passengers in the vehicle for new services of this kind. We used video-text embedding as a method for recommending multimedia and conducted experiments using movie trailers and plots. Previous studies on video-text embedding have offered new data, such as audio, or proposed new models based on high-complexity deep learning methods. In this paper, however, video-text embedding is performed based on a pretrained model and multiplication operations between features and vectors. The proposed method shows that video-text embedding can be performed without consuming much time or computing power. Similar movies and dissimilar movies can be identified through the proposed videotext embedding. This was verified by measuring genome matrix similarity, and the information about the movies used for the recommendation confirmed that the results were significant.