S-BIRD: A Novel Critical Multi-Class Imagery Dataset for Sewer Monitoring and Maintenance Systems

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Automated and robotic systems increasingly utilize computer vision for sewer maintenance.
• AI advancements enhance computer vision's capability in detecting underground sewer pipe issues like blockages and damages.
• High-quality, labeled imagery data is crucial for training effective AI detection models.

Purpose of the Study:

• Introduce the Sewer-Blockages Imagery Recognition Dataset (S-BIRD) to address the prevalent issue of sewer blockages.
• Analyze the dataset's suitability for real-time detection tasks, considering its strength, performance, consistency, and feasibility.
• Demonstrate the application of the S-BIRD dataset in an embedded vision-based robotic system for real-time sewer blockage detection and removal.

Main Methods:

• Development and presentation of the S-BIRD dataset, focusing on blockages caused by grease, plastic, and tree roots.
• Training the YOLOX object detection model using the S-BIRD dataset to validate its consistency and viability.
• Analysis of dataset parameters for real-time detection applications.

Main Results:

• The S-BIRD dataset was created to address the significant problem of sewer blockages.
• The YOLOX model demonstrated the dataset's consistency and viability for object detection.
• The study highlights the necessity of such datasets, supported by a survey in Pune, India.

Conclusions:

• The S-BIRD dataset is a valuable resource for developing AI-powered sewer maintenance solutions.
• Computer vision and AI, utilizing this dataset, can enable real-time detection and removal of sewer blockages.
• The research underscores the need for robust datasets to advance automated sewer inspection and repair.