Object Classification in Semi Structured Enviroment Using Forward-Looking Sonar.

Area of Science:

• Robotics
• Computer Vision
• Acoustic Imaging

Background:

• Increasing use of robotic systems in submarine exploration necessitates environmental understanding.
• Automated object recognition is critical for robotic tasks like monitoring and maintenance.
• Forward-Looking Sonar (FLS) provides acoustic images for underwater scene analysis.

Purpose of the Study:

• To develop and evaluate an underwater object classification pipeline for acoustic images.
• To improve the environmental perception capabilities of autonomous underwater vehicles.
• To compare the performance of different machine learning classifiers for underwater object recognition.

Main Methods:

• Object segmentation using thresholding, connected pixel searching, and intensity peak analysis.
• Feature extraction of intensity and geometric properties for object description.
• Classification using Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Trees algorithms.
• Development of an open-source tool for annotation, classification, and performance evaluation.

Main Results:

• The proposed pipeline efficiently segments and classifies structures in real-world harbor datasets.
• Experimental results demonstrate the robustness and accuracy of the developed method.
• Comparison highlights the effectiveness of the chosen segmentation and classification techniques.

Conclusions:

• The developed underwater object classification pipeline is effective for analyzing acoustic images.
• The method enhances the ability of robots to understand their environment for complex tasks.
• The open-source tool facilitates further research and development in underwater robotics and computer vision.