Fish Length Estimates from Autonomous Underwater Stereo Vision System

A non-invasive pipeline for automatic fish length estimation achieves a Mean Absolute Error of 1.30 cm and a Mean Absolute Percentage Error of 4.53%. The method uses an underwater Stereo-Vision System with YOLOv11-based segmentation and BoT-SORT tracking, evaluated on datasets of increasing complexity. The pipeline is integrated into an autonomous system for real-time on-board operation in marine environments.

Use Cases

• Training fish length estimation models based on synchronized stereo image pairs and 3D reconstruction.
• Developing non-invasive marine population assessment tools based on automated segmentation and multi-object tracking.
• Benchmarking computer vision algorithms for underwater object detection and tracking in challenging conditions.
• Calibrating autonomous underwater monitoring systems for real-time, on-board data processing.

Strengths

• The method reports specific performance metrics, including a Mean Absolute Error of 1.30 cm and a success rate of 49.7%.
• The pipeline was evaluated on two datasets of increasing complexity, including live underwater footage.
• It is designed for real-time, on-board operation in marine environments.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The overall success rate is reported as 49.7%, indicating room for improvement in detection or tracking reliability.

Provenance

Source

Caterina Muntaner-Gonzalez via figshare

Collection Method

Data likely generated from an autonomous underwater Stereo-Vision System (SVS) pipeline for 3D reconstruction and fish length estimation.

Freshness

Last updated 2026-05-25 04:33:15; freshness should be verified.