This study proposes an AI-assisted method for identifying automotive design styles, aiding students in their creative processes. Common vehicle styles in the market can be classified into two types: "family" and "sporty." Students must first understand how to classify these two design styles. This study utilizes the Faster R-CNN model to train an AI system to recognize these automotive design styles. The study employs three sample sets. The first set consists of images of gasoline cars, achieving an identification accuracy of 74.51%, with 84% accuracy for the family style and 72% for the sporty style. The second set includes images of electric cars, with an identification accuracy of 60.61%, where the family style achieved 50% accuracy and the sporty style 75%. The third set mixes images of both gasoline and electric cars for training, resulting in an identification accuracy of 68.68%, with 46.34% accuracy for the family style and 92.68% for the sporty style. The experiment yielded more promising results with gasoline car data. However, in the case of electric vehicles, since customers tend to favor the sporty design style, designers often incorporate sporty elements into family designs. This convergence of design elements makes it increasingly difficult for the recognition model to differentiate between the two.
This study demonstrates how AI can quickly identify different vehicle design styles, helping students grasp the characteristics of automotive design styles early on, and providing valuable references and support.

Authors:
Hsin-Yin Hsieh, National Cheng Kung University, Taiwan
Meng-Dar Shieh, National Cheng Kung University, Taiwan

About the Presenter(s)
Hsin-Yin Hsieh is currently pursuing a PhD in the Department of Industrial Design at National Cheng Kung University and is currently focusing on integrating AI image recognition and generative technologies into the field of industrial design.

See this presentation on the full schedule – Monday Schedule

Bookmark