[Paper Review] Non-invasive Growth Monitoring of Small Freshwater Fish in Home Aquariums via Stereo Vision

Clemens Seibold, Anna Hilsmann+1 more|arXiv (Cornell University)|Mar 6, 2026

Water Quality Monitoring Technologies0 citations

TL;DR

A stereo-camera system using refraction-aware epipolar geometry and a YOLOv11-Pose network estimates 3D fish keypoints to measure length non-invasively in aquariums. It includes a learned quality filter and template-based keypoint refinement to improve accuracy.

ABSTRACT

Monitoring fish growth behavior provides relevant information about fish health in aquaculture and home aquariums. Yet, monitoring fish sizes poses different challenges, as fish are small and subject to strong refractive distortions in aquarium environments. Image-based measurement offers a practical, non-invasive alternative that allows frequent monitoring without disturbing the fish. In this paper, we propose a non-invasive refraction-aware stereo vision method to estimate fish length in aquariums. Our approach uses a YOLOv11-Pose network to detect fish and predict anatomical keypoints on the fish in each stereo image. A refraction-aware epipolar constraint accounting for the air-glass-water interfaces enables robust matching, and unreliable detections are removed using a learned quality score. A subsequent refraction-aware 3D triangulation recovers 3D keypoints, from which fish length is measured. We validate our approach on a new stereo dataset of endangered Sulawesi ricefish captured under aquarium-like conditions and demonstrate that filtering low-quality detections is essential for accurate length estimation. The proposed system offers a simple and practical solution for non-invasive growth monitoring and can be easily applied in home aquariums.

Motivation & Objective

• Motivate non-invasive, frequent growth monitoring of small freshwater fish to assess welfare, health, and nutrition in home aquariums.
• Develop a stereo vision system that accounts for air–glass–water refraction to accurately estimate fish length.
• Create and share a stereo image dataset of Sulawesi ricefish with annotations for bounding boxes, keypoints, and quality.
• Evaluate the impact of quality assessment and filtering on matching and length estimation accuracy.

Proposed method

• Extend YOLOv11-Pose to detect fish, predict five anatomical keypoints, and estimate an overall image-quality level.
• Apply refraction-aware epipolar constraints and 3D triangulation to estimate fish 3D keypoints from stereo pairs.
• Use a three-term stereo matching cost that combines epipolar curve proximity, bounding-box size similarity, and keypoint configuration.
• Refine keypoints with template matching constrained along epipolar curves.
• Filter unreliable matches using quality predictions and orientation-based checks to improve length estimation robustness.

Experimental results

Research questions

• RQ1 Can a refraction-aware stereo pipeline accurately estimate the length of small freshwater fish in aquarium settings?
• RQ2 Does integrating a learned quality estimator and orientation filters reduce erroneous stereo matches and improve length RMSE?
• RQ3 How does background context (wall vs. plants) affect template-matching improvements in length estimation?
• RQ4 What is the runtime performance of the proposed system on standard hardware for practical home aquarium use?

Key findings

• The system achieves non-invasive fish length estimation by triangulating refraction-aware 3D keypoints from stereo images.
• Quality filtering and orientation checks significantly reduce incorrect matches across YOLO backbones, with best results using medium backbone and quality+direction filters.
• Template matching for keypoint refinement can improve RMSE in wall-background scenes but may worsen it in cluttered scenes, indicating background-dependent effects.
• Across test scenes, the approach runs at about five frame pairs per second on standard hardware, with template matching contributing ~75% of runtime.
• RMSE improvements of up to about 50% are reported for certain configurations, particularly when applying both quality filtering and direction-based filtering.