딥러닝 기반 객체 인식을 이용한 AR파이프 모델의 자동 텍스쳐링 방법에 관한 연구

초록

Augmented reality models used in the shipbuilding industry generate and map textures using manual work or expensive equipment for photo-realistic textures. However, if there is a way to learn and create similar textures to actual models and apply them to virtual models, productivity can be increased by reducing the production stage of augmented reality models. To this end, the present study carried out a study using Cyclegan to generate textures that have learned the actual model and to automatically map it to an augmented reality model. Cyclegan is a GAN-based algorithm for image translation that transfer only certain parts of the image into desired forms. However, existing CycleGAN cannot be applied to complex types of pipes, so algorithms combined with object Recognition were developed and applied to the industry. The findings are considered to be the basis of the research that effectively produces textures in the future and prove that actual textures can be applied to augmented reality models.