PROCEEDINGS OF THE 27TH INTERNATIONAL ACM SIGACCESS CONFERENCE ON COMPUTERS AND ACCESSIBILITY, ASSETS 2025(2025)
Nihon Univ
被引用0|浏览0
摘要
Improperly installed tactile paving may cause confusion or accidents for visually impaired pedestrians. This preliminary study investigates an automated approach to identifying such installation errors. We adopt DeepLabV3+, a semantic segmentation model, to distinguish between two types of tactile blocks-warning and guiding blocks-thereby facilitating visual assessment of improper layouts by human observers. The model achieved a mean Intersection over Union (mIoU) of 0.937 +/- 0.005 across 5 cross-validation folds, demonstrating promising segmentation performance. We further discuss the effectiveness of this segmentation-based approach in enhancing the detectability of improper installations.