Session: 05-04-01 CCUS and Underwater Development/ Utilization

Submission Number: 155491

Enhancing Autonomous Recognition in Deep-Sea Environments With AI-Based Target Detection and 3D Depth Estimation 

The authors aim to develop an automatic target detection system that enables high-precision target detection and tracking in deep-sea environments for 11,000-meter-class underwater vehicles.
So far, the detection model DETR and the tracking algorithm Byte-Track have been implemented, and during demonstration tests from June to July 2023, real-time target detection and tracking were successfully achieved. However, issues such as decreased detection accuracy when objects are partially hidden or overlapped, and the differences between training data and real-world conditions, have become apparent.
To address these issues, the authors added newly collected ocean trial data and data of hidden targets to the training process, creating a new model that demonstrated superior detection accuracy. Additionally, by optimizing the parameters of the tracking algorithm, the stability of tracking was enhanced, and AI-based depth estimation (CREStereo) was introduced to improve the accuracy of 3D measurements. This method was tested in a demonstration in October 2024, and the disparity map creation method was improved compared to previous approaches.
These improvements have increased the reliability of real-time automatic sampling tasks in deep-sea environments, marking an important step toward practical application.
However, there are still challenges in detecting unknown or overlapping objects, which will need to be addressed in future research. Furthermore, AI-based depth estimation still requires some improvement in processing speed, and further optimization is needed to achieve complete real-time processing.

Presenting Author: Kazuya Iwashita Japan Agency for Marine-Earth Science and Technology

Presenting Author Biography: Engineer, Japan Agency for Marine-Earth Science and Technology

B.Sc. in Mathematics, Faculty of Science, Chiba University

He is currently focused on the research and development of machine learning models for image recognition in underwater exploration.