3D profilometric object detection in turbid water using integral imaging and deep neural networks

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Oceanography

Background:

• Underwater object detection is crucial for marine exploration and operations.
• Turbid water conditions significantly degrade the performance of conventional imaging systems.
• Integral Imaging (InIm) offers potential for 3D scene reconstruction but faces challenges in turbid environments.

Purpose of the Study:

• To develop and evaluate a robust 3D object detection system for turbid water environments.
• To leverage 3D profilometry and deep neural networks for enhanced underwater perception.
• To demonstrate the superiority of the proposed method over existing 2D and 3D imaging techniques.

Main Methods:

• A deep neural network-based framework for red, green, blue-depth (RGB-D) object detection was developed.
• Passive 3D profilometry was employed, capturing multiple 2D perspective images from a moving platform.
• A depth map was statistically estimated and fused with perspective images to create a four-channel RGB-D image for detection.

Main Results:

• The proposed 3D profilometry-based approach demonstrated superior performance in object detection within turbid water.
• The system outperformed both traditional 2D imaging and conventional 3D InIm reconstruction methods.
• Effectiveness was validated across various tested levels of water turbidity.

Conclusions:

• The novel InIm 3D profilometry system offers a significant advancement for object detection in challenging turbid underwater conditions.
• This research presents the first application of InIm 3D profilometry for object detection specifically in turbid water.
• The developed RGB-D framework provides a robust solution for underwater autonomous systems.