WEKO3
アイテム
A Study on Road Surface Marking Quality Evaluation Using Machine Learning and Computer Vision
http://hdl.handle.net/10076/0002000907
http://hdl.handle.net/10076/000200090771a89b08-649f-4b5b-bd5e-3c214eeed02d
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
2024DE0702.pdf (33.8 MB)
|
|
| Item type | 学位論文 / Thesis or Dissertation(1) | |||||||
|---|---|---|---|---|---|---|---|---|
| 公開日 | 2024-10-09 | |||||||
| タイトル | ||||||||
| タイトル | A Study on Road Surface Marking Quality Evaluation Using Machine Learning and Computer Vision | |||||||
| 言語 | en | |||||||
| 言語 | ||||||||
| 言語 | eng | |||||||
| 資源タイプ | ||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_db06 | |||||||
| 資源タイプ | doctoral thesis | |||||||
| アクセス権 | ||||||||
| アクセス権 | open access | |||||||
| アクセス権URI | http://purl.org/coar/access_right/c_abf2 | |||||||
| 著者 |
Boudissa, Mehieddine
× Boudissa, Mehieddine
|
|||||||
| 抄録 | ||||||||
| 内容記述タイプ | Abstract | |||||||
| 内容記述 | In the context of rapidly accelerating urban expansion and advancements in deep learning, the need for sustainable and efficient traffic management systems has emerged as a pivotal research area. This dissertation addresses the challenges associated with the deterioration of road surface markings, particularly in the Mie prefecture of Japan. Central to this research is a comprehensive dataset of 13,000 high-resolution images (3000x1600) sourced from the local government facilities of Mie prefecture. These images span diverse roads and intersections in urban, rural, and even off-road contexts. Captured under varied lighting conditions, they present a slew of challenging samples, including images with glare effects, shadowy regions, deteriorated roads, and traffic signs. Through the development of a semiassisted annotation tool, an initial subset of 400 images was labeled, serving to train a U-Net model that achieved a Dice score of 78.90%. Recognizing the potential of the extensive dataset, a subsequent phase of assisted labeling was undertaken. This effort used the trained model’s inference on all images, facilitating a streamlined labeling process. Ultimately, this method yielded 12,000 labeled images, with about 1,000 images deemed unfit for accurate annotations. The research’s early stages successfully detected and segmented these landmarks, integrating both classical computer vision techniques and deep learning approaches. The introduction of the ”Efficient VGG-16” model, a tailored version of the renowned VGG-16, emerged as a significant contribution, adeptly evaluating road surface marking quality and achieving a Mean Squared Error (MSE) of 3.62. Further deepening the research, a comprehensive survey of various segmentation models was conducted, with the U-Net model exhibiting notable superiority in terms of Dice score evaluation. The current trajectory of this research is marked by several promising endeavors: The exploration of uncertainty-aware regression to refine road surface marking quality evaluation. The application of Diffusion techniques to enhance road surface marking quality assessment. The integration of PhyCV edge detection for a holistic approach to road surface marking detection and evaluation. Conclusively, this dissertation underscores the transformative potential of advanced computer vision techniques in the realm of traffic management and road safety, heralding a new era of research and practical implementations in the domain. |
|||||||
| 言語 | en | |||||||
| 内容記述 | ||||||||
| 内容記述タイプ | Other | |||||||
| 内容記述 | 本文/Division of Systems Engineering, Graduate School of Engineering, Mie University | |||||||
| 内容記述 | ||||||||
| 内容記述タイプ | Other | |||||||
| 内容記述 | 125p | |||||||
| 書誌情報 |
発行日 2024-07-17 |
|||||||
| フォーマット | ||||||||
| 内容記述タイプ | Other | |||||||
| 内容記述 | application/pdf | |||||||
| 著者版フラグ | ||||||||
| 出版タイプ | VoR | |||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||
| 出版者 | ||||||||
| 出版者 | 三重大学 | |||||||
| 出版者（ヨミ） | ||||||||
| 値 | ミエダイガク | |||||||
| 学位名 | ||||||||
| 学位名 | 博士(工学) | |||||||
| 学位授与機関 | ||||||||
| 学位授与機関識別子Scheme | kakenhi | |||||||
| 学位授与機関識別子 | 14101 | |||||||
| 学位授与機関名 | 三重大学 | |||||||
| 学位授与年月日 | ||||||||
| 学位授与年月日 | 2024-07-17 | |||||||
| 学位授与番号 | ||||||||
| 学位授与番号 | 甲工学第2275号 | |||||||
| 資源タイプ（三重大） | ||||||||
| 値 | Doctoral Dissertation / 博士論文 | |||||||
