Tactile Paving Detection and Classification Method Based on Cyclist-Participatory Road Image Sensing

Yuto Matsuda, Yuki Matsuda: “Tactile Paving Detection and Classification Method Based on Cyclist-Participatory Road Image Sensing,” 27th International Conference on Distributed Computing and Networking (ICDCN '26 Companion), pp.78-83, 2026.

Abstract

This study proposes a method for collecting tactile paving location information by acquiring road surface images and GPS data using a bicycle equipped with a compact camera and a GPS module. As a preliminary experiment, road surface images were captured under different camera positions and angles to identify optimal installation conditions. An object detection model based on YOLO11 achieved tactile paving detection with a mAP_50 of 0.777. Subsequently, a Convolutional Neural Network (CNN) based on ResNet18 classified tactile paving types (guiding or warning) with a macro-F1 score of 0.898. These results demonstrate the feasibility of the approach while highlighting challenges such as model optimization for camera placement and expanding training data.

Links

DOI: https://doi.org/10.1145/3737611.3776614

PDF: https://cocolab.jp/publication/files/202601_ICDCN_YutoMatsuda.pdf

BibTeX

code:references.bib

@inproceedings{bib:yuto_brailoop_ICDCN2026,

author={Matsuda, Yuto and Matsuda, Yuki},

title={Tactile Paving Detection and Classification Method Based on Cyclist-Participatory Road Image Sensing},

booktitle={27th International Conference on Distributed Computing and Networking (ICDCN '26 Companion)},

pages={78--83},

year={2026},

doi={10.1145/3737611.3776614},

url={https://doi.org/10.1145/3737611.3776614}

}

https://scrapbox.io/files/695ab204172d0ce88db8e543.png