Underwater Object Recognition Using Point-Features, Bayesian Estimation and Semantic Information | JoVE Visualize

Underwater Object Recognition Using Point-Features, Bayesian Estimation and Semantic Information

Khadidja Himri¹, Pere Ridao¹, Nuno Gracias¹

¹Underwater Robotics Research Center (CIRS), Computer Vision and Robotics Institute (VICOROB), University of Girona, Parc Científic i Tecnològic UdG C/Pic de Peguera 13, 17003 Girona, Spain.

Sensors (Basel, Switzerland)|April 3, 2021

View abstract on PubMed

Summary

This study presents a 3D object recognition method for sub-sea industrial structures using point features. Incorporating Bayesian estimation and semantic information significantly improves recognition accuracy for underwater inspection, maintenance, and repair (IMR).

Area of Science:

Robotics
Computer Vision
3D Object Recognition

Background:

Industrial sub-sea structures require robust Inspection, Maintenance, and Repair (IMR).
Accurate 3D object recognition from non-coloured point clouds is crucial for sub-sea IMR.
Existing methods may lack the precision needed for complex underwater environments.

Purpose of the Study:

To propose and evaluate a novel 3D object recognition method for non-coloured point clouds.
To enhance the accuracy of object recognition for sub-sea industrial structures.
To investigate the impact of feature descriptors, Bayesian estimation, and semantic information on recognition performance.

Main Methods:

A 5-stage recognition pipeline including plane segmentation, pipe detection, semantic object-segmentation, feature-based recognition, and Bayesian estimation.
Development of an Interdistance Joint Compatibility Branch and Bound (IJCBB) algorithm for object tracking and Bayesian estimation.
Utilizing the Clustered Viewpoint Feature Histogram (CVFH) descriptor for point feature extraction.

Main Results:

The proposed method achieves high recognition rates, reaching up to 90% with optimal settings.
Bayesian estimation improved recognition rates by 18%.
Inclusion of semantic information further increased recognition rates by an additional 21%.

Conclusions:

The CVFH descriptor, combined with Bayesian estimation and semantic information, provides a highly effective 3D object recognition solution for sub-sea IMR.
The proposed method demonstrates significant performance gains over traditional approaches.
This technique is well-suited for autonomous underwater vehicle (AUV) navigation and data analysis in sub-sea environments.