Remote drain inspection framework using the convolutional neural network and re-configurable robot Raptor.

Area of Science:

• Environmental Engineering
• Robotics
• Computer Vision

Background:

• Drain blockages pose significant risks to urban ecosystems and public health.
• Manual drain inspection methods are hazardous, inefficient, and expose workers to diseases.
• Automated inspection is crucial for maintaining urban infrastructure and environmental safety.

Purpose of the Study:

• To develop and evaluate an automated drain inspection framework.
• To enhance the safety and efficiency of urban drain maintenance.
• To accurately detect and classify drain-blocking objects using AI.

Main Methods:

• A convolutional neural network (CNN) based object detection algorithm was employed.
• Transfer learning was utilized to train the CNN model on a custom dataset of drain-blocking objects.
• A custom-built teleoperated robot, 'Raptor', was developed for drain navigation and inspection.

Main Results:

• The CNN object detection model achieved 91.42% accuracy in detecting and classifying drain-blocking objects.
• The system demonstrated a processing speed of 18 frames per second (FPS).
• Field trials confirmed the 'Raptor' robot's stable maneuverability, accurate mapping, and localization in complex drain environments.

Conclusions:

• The developed CNN-based framework and 'Raptor' robot offer an effective solution for automated drain inspection.
• This technology significantly improves the accuracy and efficiency of identifying drain blockages.
• The system enhances worker safety and contributes to better urban environmental management.