Accurate estimation of fish length in single camera photogrammetry with a fiducial marker

Abstract

Videogrammetry and photogrammetry are increasingly being used in marine science for unsupervised data collection. The camera systems used to collect such data are varied and often complex. In contrast, digital cameras and smartphones are ubiquitous, convenient for the user and an image automatically captures much of the data normally recorded on paper as metadata. The limitations of such an approach are primarily attributed to errors introduced through both the image acquisition process and lens distortion of the collected images. In the present study, a methodology is presented to achieve accurate 2-dimensional (2-D) total length (TL) estimates of fish without specialist camera equipment or proprietary software. Photographs of depressed (flat) and fusiform fish were captured with an action camera using a background fiducial marker, positioned at the distal plane of the subject; a foreground fiducial marker, at the proximal plane of the subject and a laser marker, projected on to the subject's surface. To correct image distortions, the geometric properties of the lens were modelled with OpenCV. The accuracy of TL estimates were corrected for parallax effects using a novel iterative algorithm requiring only the initial length estimate and known morphometric relationships. OpenCV was effective in correcting image distortion, decreasing RMSE by 96% and the percentage mean bias error (%MBE) by 50%. By undistorting the image and correcting for parallax effects a % MBE [95% CIs] of-0.6% [-1.0,-0.3] was achieved and RMSE was reduced by 86% to 2.1%. Estimation of the lens to subject distance using the similar triangles calibration method resulted in the best estimation of TL. The present study demonstrates that the morphometric measurement of different species can be accurately estimated without the need for expensive, complex or bulky camera equipment making it particularly suitable for deployment in citizen science and other volunteer based data collection endeavours.

Smart Artificial Markers for Accurate Visual Mapping and Localization

“…This type of application relies on highly-controlled images acquired from specific cameras in general with a high-precision capture process—with a high associated cost, however. Some of the techniques can be of inspiration and even adapted for lower-cost equipment such as cell phones and cameras, and the community has done it since the appearance and first uses of digital photography [ 8 , 32 ]. Nonetheless, most of the models cannot be straightforwardly applied in real-time applications such as robotics and virtual reality.…”

AragoJ: A free, open‐source software to aid single camera photogrammetry studies

“…Nonetheless, the method has been shown to be both accurate and to produce suitable levels of precision, when properly implemented (e.g. Dawson et al, ; Deakos, ; Galbany et al, ; Grochowsky, Alaways, Siskey, Most, & Kurtz, ; Jadejaroen, Hamada, Kawamoto, & Malaivijitnond, ; Jeffreys, Rowat, Marshall, & Brooks, ; Monkman, Hyder, Kaiser, & Vidal, ; Weisgerber, Medill, & McLoughlin, ).…”

Deep learning based fish length estimation. An application for the Mediterranean aquaculture

“…There exist optical methods tackling the problem of biomass estimation applying stereoscopy (e.g (Harvey et al, 2003) for individual measurements while there are some exceptions that fish body length is estimated by one camera (Monkman et al, 2020). The proposed method uses convolutional neural networks (CNN) which have proven successful for the task (Cao et al, 2019).…”