Visual pollution real images benchmark dataset on the public roads

Area of Science:

• Environmental Science
• Urban Planning
• Computer Vision

Background:

• Quality of Life (QoL) is multifaceted, with environmental quality being a key component.
• Visual Pollution (VP) negatively impacts urban environments, specifically affecting public roads with elements like excavation barriers, potholes, and dilapidated sidewalks.
• Automated detection of VP is challenging due to its subjective nature and lack of standardized assessment methods.

Purpose of the Study:

• To develop a comprehensive dataset for visual pollution detection in Saudi Arabia.
• To facilitate the automatic prediction and detection of visual pollution by government agencies.
• To enhance the research domain with a publicly available VP image dataset.

Main Methods:

• A dataset of 34,460 RGB images was collected from various regions in the Kingdom of Saudi Arabia (KSA).
• Images were categorized into three classes: excavation barriers, potholes, and dilapidated sidewalks.
• Deep Active Learning (DAL) strategy was employed for image annotation, with initial manual annotation by four experts.

Main Results:

• The dataset includes 34,460 images, annotated for detection and classification tasks.
• The dataset contains 8,417 bounding boxes for excavation barriers, 25,975 for potholes, and 7,412 for dilapidated sidewalks.
• The MOMRAH dataset is now publicly available.

Conclusions:

• The developed MOMRAH dataset provides a valuable resource for research in visual pollution detection.
• Automated VP detection systems can be developed using this dataset to improve urban landscapes.
• This initiative supports Saudi Arabia's campaign to enhance its urban environment.