Citation

Iwasaki H, Hayashi H, Kamezaki M, Sugano S. Trans. Soc. Automot. Eng. Jpn. 2023; 54(3): 528-533.

Vernacular Title

Image Inpainting を用いたハンドル映り込み画像における運転者姿勢推定の高精度化

Copyright

(Copyright © 2023, Society of Automotive Engineers of Japan)

DOI

10.11351/jsaeronbun.54.528

PMID

unavailable

Abstract

The camera that monitors bus drivers' glance behavior and pose must be placed behind the steering-wheel to avoid obstructing driver's view. The problem is that the steering-wheel occludes part of driver's body, which would reduce the accuracy of driver's pose estimation. In this study, we propose an approach to estimate the steering-wheel area by using semantic segmentation and remove this area from the image by using Image Inpainting. Our experimental results showed that the accuracy of pose estimation could be improved and using Pluralistic Image Completion (PIC) for removing the steering wheel had the best results. In addition, the experimental results showed that adjusting the variable z of PIC before driving can detect posture with higher accuracy.

===

バス運転者の視線や姿勢をモニタリングするカメラは，視界の妨げにならないようハンドルの後ろに配置せざるを得ない．ハンドルが体の一部を遮るため，姿勢推定の精度低下が問題となる．本研究ではImage Inpaint技術を活用し，画像からハンドルを認識・除去・修復する手法を提案する．実験により姿勢推定の精度向上を確認できた．

Language: ja

Keywords

Driver position; Driving posture; Human engineering; Inpainting; Semantic segmentation