Presentation Information

[NpVBMPAo-10]Proposal of Quantitative Deterioration Assessment Method for Pavement Markings using Semantic Segmentation

*Nobuyuki Oe1, Tomonori Aihara1, Hisashi Takagi1, Tomoki Tajima1, Tomoaki Sakai1 (1. Kokusai Kogyo Co., Ltd (Japan))

Keywords:

Pavement Marking,Autonomous Driving,Semantic Segmentation,Deterioration Assessment,Artificial Intelligence (AI)

In Japan, as the roadmap for the advancement and practical realization of autonomous driving progresses, the demand for maintaining pavement markings, which serve as indicators for self-localization, is rapidly increasing. However, the inspection and evaluation of pavement markings currently lag behind that of road structures. The primary factor for this delay is that, compared to structural damage such as pavement cracks, pavement marking deterioration is often viewed as having lower urgency. Consequently, it is difficult to justify the cost-effectiveness of conducting standalone inspections, leading to a lack of accumulated stock data over wide areas.
To address this issue, this study aims to develop an image analysis model capable of quantitatively analyzing the peeling degree of pavement markings using road images acquired by patrol cameras or commercial dashcams. Therefore, this study initially limits its scope to longitudinal pavement markings (left and right lines), prioritizing these markings as they constitute the vast majority of road marking infrastructure and play a critical role in lane keeping.
The proposed method applies Semantic Segmentation—an AI analysis technique—to road images obtained from vehicle-mounted cameras. This enables the pixel-by-pixel identification of pavement marking positions and peeling areas.
The establishment of this technology achieves both the quantitative calculation of peeling rates (%) and the determination of the 5-stage peeling rank proposed by the Japan Road Signs and Pavement Markings Association based on pixel information. Furthermore, since the method utilizes detachable, aftermarket equipment rather than dedicated survey vehicles, it enables simultaneous execution with other surveys (e.g., pavement condition surveys). This resolves the issues of cost and efficiency associated with standalone inspections, thereby realizing efficient and quantitative inspection over large distances and contributing to the advancement of road maintenance management.Future work will focus on expanding the scope to various pavement markings necessary for autonomous driving, such as stop lines and crosswalks, to develop technologies applicable to comprehensive road maintenance.