Title Intelligent image recognition analyses for wave overtopping on coastal highways and seawalls (2/4) - Image recognition of wave overtopping in the nighttime
Dept Transportation Technology Research Center
Year 113
Month 3
Price 200
Summary     To reduce the disasters caused by overtopping and wave attacks on coastal roads and port areas, this research project aims to analyze the areas prone to overtopping by utilizing data from a CCTV system, in conjunction with image recognition techniques. The project focuses on two areas: the "Provincial Highway No. 11 " and the "Hualien Port." Through the established camera systems,the project involves collecting image data and analyzing the rise and overtopping of waves, and the findings will be reported to the Institute of Transportation, Ministry of Transportation and Communications, to serve as a basis for overtopping warnings.
    This four-year project is currently in its second year (2023). Building upon the first year's work of "developing image recognition technology for tracking overtopping during the day," this year's main goal is to complete nighttime overtopping image interpretation techniques. Moreover, the project includes developing a rise/overtopping model for the Provincial Highway No. 11 using machine learning, incorporating data from the Taiwan Coastal Operational Modeling System 2.0 (TaiCOMS2.0), to estimate the rise on the highway, and comparing these results with existing coastal road wave attack early warning systems to evaluate their effectiveness.
    Based on the achievements of the previous year, a day/night image interpretation system for tracking rises has been established in the Provincial Highway No. 11. By deploying optical and thermal imaging equipment, the project successfully collected daytime and nighttime images during typhoons such as MAWAR, DOKSURI, KHANUN, SAOLA, and HAIKUI. Thermal imaging, which captures changes in thermal temperatures, compensates for the limitations of optical imaging at night and is effective in capturing rise images, especially during the early stages of rainfall when the temperature difference remains significant. However, thermal imaging for waterline interpretation at night requires a noticeable temperature difference for effective recognition, and is not always reliable for warning purposes. To make image interpretation technology applicable for coastal road wave attack alerts, an automated image interpretation rise system has been established this year. Based on the analysis of two years' worth of images, a threshold value for image waterline recognition has been set to ensure reliable results in automated interpretation. Additionally, this automation has been integrated with an image-based warning line judgment system to issue overtopping warning signals for practical use in road wave attack alerts.
    Furthermore, following the advanced planning results of the existing wave attack warning system from the initial project phase, this year's work involves strengthening the warning system. By employing machine learning methods (Gaussian Process), a relationship between offshore sea conditions and nearshore rise (including wave direction considerations) has been established for the Ren Ding Sheng Tian section. With the input of TaiCOMS2.0 ocean forecast data, successful rise forecasting is now possible. By comparing these forecast results with image interpretation and sensor data, it is found that while there are some discrepancies between image analysis and rise forecasting compared to sensor observation data, they generally fall within an acceptable error margin and can serve as a reference for future refinement of forecast models and image analysis.
Post date 2024/03/25
View count: 223