Marine Mammal Morphometrics: Computer Vision-Assisted Photogrammetry, Three-Dimensional Modeling, and Validations

Abstract

Marine mammal morphometrics contribute to our understanding of their physiology, biomechanics, behavior, ecology, and conservation. The morphometric variables of body dimensions, volume, and mass are challenging to obtain, especially for cetaceans, which are often large and inaccessible. Current morphometric data come primarily from whaling, bycatch, strandings, and captive individuals. The challenge of obtaining reliable morphometric data from free-ranging individuals has improved using high-resolution digital imaging and three-dimensional modeling (3DM). With the development of unmanned (or unoccupied) aerial vehicles (UAVs, often abbreviated as "drones"), images can be used to create accurate 3D models of individuals. These models can be scaled to obtain morphometric variables, which can be used in modeling body volume and mass. Although several pioneering studies have attempted 3D models with cetaceans, the technique of image-oriented 3D models has not been validated. All mathematical models have potential errors, so 3D models are more useful when validated with direct measurements.

For image-oriented modeling techniques, the automation of conventional image processing increases efficiency, especially when many images are processed. Computer vision is becoming integrated into artificial intelligence (AI), enabling automated image processing that is otherwise labor-intensive and time-consuming. Here I present a practical and efficient program for obtaining morphometrics in a fully automated model of the East Asian finless porpoise (Neophocaena asiaeorientalis sunameri, hereafter referred to as EAFP). As one of the smallest cetaceans, the EAFP occurs in the East China Sea, the Yellow Sea, and the seas around Japan. This small odontocete is commonly observed in shallow water (up to 50 m in depth) along the shore and in estuaries and is thus susceptible to anthropogenic threats. The estimated population is approximately 1,800, making EAFP an endangered species (IUCN 2015). The maximum recorded body length and mass from my data were 227 cm and 72 kg, respectively. I took advantage of the small size of EAFP to measure their total body volume (TBV) and mass (BM) and used these data to validate 3D models of these variables.

In this study, I validated 3D modeling as an accurate method to measure morphometrics. I showed that the conventional truncated cone and elliptical cone models were susceptible to significant error depending on the number of transverse sections. Computer vision-assisted photogrammetry fully automated image processing and was 385-fold faster than non-automated photogrammetry. In addition, I showed that a 3D model of TBV based on one dorsal image achieved 63% accuracy relative to models using two perspectives. My results will be helpful to other investigators who wish to obtain estimates of TBV and BM for various cetacean species using images from UAVs and computer vision-assisted photogrammetry and 3D modeling.