In a paper published in the journal Scientific Reports, researchers presented a novel algorithm to monitor severe weather conditions accurately to mitigate traffic accidents on highways. The algorithm combined frequency and spatial domain techniques to construct a rain coefficient model, enabling the identification of rainy days and assessment of rainfall intensity.
Validation using data from different weather conditions on the Jinan bypass G2001 line demonstrated the model's effectiveness in distinguishing between weather conditions and evaluating rainfall intensity, with implications for enhancing road traffic safety.
Related Work
Past work has primarily focused on using machine learning (ML) or deep learning (DL) methods for severe weather research, albeit with limitations such as difficulty accurately judging rainy days and high computational requirements. These methods have shown promise but need help accurately discerning rainy conditions due to the complexity of weather patterns and variations in image quality. Additionally, the high computational demands of ML and DL approaches pose constraints, especially in resource-limited environments. Addressing these issues is crucial for developing robust and efficient rainy-day recognition systems for enhanced road safety.
Rainy Day Monitoring Model Development
The study area selected for this research is the region captured by the camera positioned at K15+0.10 km on Jinan Bypass Highway G2001, a vital link in the regional transportation network connecting major national highways in Jinan, Shandong Province. The area's transportation system is pivotal in facilitating regional travel.
The study collected video data from high-definition surveillance equipment stationed at the meteorological observatory in Jinan between September and October 2022. Videos captured during different weather conditions—sunny, cloudy, and rainy days—were analyzed to understand the corresponding weather conditions for each video section.
Researchers examined grayscale histograms of images from sunny, cloudy, and rainy days to analyze the characteristics of various weather conditions. Grayscale histograms revealed distinctive features for each weather condition, with sunny days displaying stable histograms. In contrast, on cloudy and rainy days, they exhibited double-peak patterns. Rainy days showed an additional peak attributed to water accumulation on roads due to rainfall.
The construction of the rainy day monitoring model relied on several principles. These included employing the optimal thresholding using Saikatsu's method (OTSU) method for adaptive threshold segmentation, high-pass filtering to extract contours and remove road and vehicle effects, global threshold segmentation to binarize images, masking to isolate areas of interest, and morphological processing to eliminate noise.
The researchers constructed the rainy day monitoring coefficient model using image processing techniques and algorithms based on these principles. The model calculated the proportion of white pixel blocks in road sections to distinguish between weather conditions and assess rainfall intensity. Each method, including the OTSU method, high-pass filtering, complete domain value segmentation, masking, and morphological processing, played a crucial role in model construction, ultimately contributing to accurate rainy-day highway monitoring.
Active Analysis of Weather Data
The study area selected for this research encompasses the region captured by a camera at K15+0.10 km on Jinan Bypass Highway G2001, a crucial link in Jinan's regional transportation network. Researchers analyzed video data collected between September and October 2022 from high-definition surveillance equipment stationed at the meteorological observatory in Jinan to characterize various weather conditions, including sunny, cloudy, and rainy days.
The study processed camera data using Python3.10, PyCharm2022.2.2, and OpenCV4.6.0 and applied the developed monitoring model to assess rainy conditions. The Prain coefficient was calculated for different weather conditions, indicating the proportion of white pixel blocks on the road surface. The model effectively categorized weather conditions based on the Prain coefficient, distinguishing between cloudy, light rainy, moderate rainy, and heavy rainy days. Additionally, the study observed a positive correlation between rainfall intensity and road brightness due to headlight reflections.
Researchers conducted data verification using cameras at different times on the Jinan Ring Expressway, demonstrating the accuracy and applicability of the monitoring model. They extended the validation to other road sections to confirm the model's universality. Model error analysis revealed the importance of vehicle removal, masking, and morphological opening denoising operations in ensuring accuracy.
The study employed image processing techniques to construct a rain coefficient model, minimizing the influence of vehicles and noise to enhance accuracy. However, due to challenges in processing nighttime data, the model's applicability is limited to daytime rain monitoring. Validation of nighttime data demonstrated reduced accuracy, highlighting the model's limitations in low-light conditions.
Conclusion
To sum up, this paper introduced a rain monitoring model based on various image processing techniques, successfully identified rainy days, and assessed rainfall intensity. The model accurately distinguished between cloudy and rainy days, with results aligning with meteorological forecasts. While applicable to expressways and ordinary roads, limitations existed for nighttime monitoring due to reliance on grayscale values. Validation confirmed the model's accuracy, showcasing its effectiveness without vehicle removal, masking, or morphological opening denoising—future research aimed to address these limitations and integrate deep learning to enhance the model's capabilities.