Can I read Development of a Semantic Scene Conversion Model for Image-based Localization at Night on EtoBox?

Development of a Semantic Scene Conversion Model for Image-based Localization at Night by Kim, Dongyoun ;Shin, Sangwon ;Park, Jinwoo ;Shin, Sung is a scholarly article available to read on EtoBox.

What is Development of a Semantic Scene Conversion Model for Image-based Localization at Night about?

Illumination change is one of the challenges in image-based localization in smart cars application. To deal with illumination change, image conversion methods have been researched. However, these methods would lose the detail of objects in images. In this paper, we propose the Semantic Local Image Conversion (SLIC) model changing the appearance of local semantic objects in an image by categories at night. This enables the proposed model not to lose the detail of static objects in image conversion. As a result, it is expected that the proposed SLIC method has a better result in image-based localization. SLIC method uses static objects (i.e., traffic signs and street lamps) as categories for localization. The SLIC method is composed of two phases as (1) instance segmentation and (2) static objects conversion. Instance segmentation is utilized as a detector for static objects. In the conversion phase, the detected static objects are converted from the appearance of objects at night to objects at day. We then compare the visual inspection and the number of matching of converted objects with existed models (Pix2Pix with global pixels in the image and ToDayGAN). Overall, our model shows

Author

Kim, Dongyoun ;Shin, Sangwon ;Park, Jinwoo ;Shin, Sung

Publisher

ACM Press

Published

2019

Language

EN