Palaeontology has revolutionised in recent years by the use of powerful methods for the digital visualisation and analysis of fossil material, including radical improvements and increased avaliability of computer technology and tomographic techniques, enabling 3D reconstructions of specimens to be made and poorly understood or unknown details of fossil taxa, as well as microfossils or trace fossils, to be described in greater detail.
Among the major advances enabled by digitisation include the observation of structures that were near-impossible to visualise, enabling inferences of behaviour (e.g. sensory and locomotory capabilities of extinct organisms) to be made and the developmental sequence of fossil organisms to be studied by observing preserved growth lines without destructive histological sampling.
However, an area where it has become particularly important is analyses of functional morphology of fossil organisms as tools used for recreating and analysing anatomical structures become more powerful and easier to use, enabling previously untestable hypotheses to be tested without damaging potentially fragile specimens. The digitisation of skeletons has enabled researchers to analyse musculoskeletal functions and aspects of biomechanics without the imitations of having to handle large, heavy and often surprisingly fragile bones, and allowing simulations that would be impossible if relying on specimens alone.
As digital datasets proliferate, 3D data can be shared and disseminated, providing a possible solution where fossil material is rare or inaccessible and enabling for further collaborative analysis (and to safeguard valuable specimens. Digitisation is done in three main ways - laser scanning, CT scanning and photogrammetry.
Whilst laser scanners come in a variety of models, they're generally suited to a specific range and object size. In addition, they're also quite expensive - whilst they're becoming more affordable as their use becomes more prevalent, not many palaeontology research groups own their own scanners. In addition, once data has been acquired, proprietary software and/or a high level of expertise is necessary to both align individual scans and clean spurious data. For my study, I used photogrammetry as a digitisation tool.
What is photogrammetry? And how did I use it in my study?
Photogrammetry is a digitisation method using overlapping photographs of an object taken with a digital camera to create a 3D digital model. Compared to laser scanning, photogrammetry is more accessible - you only need a camera and computer - and is more affordable, with workflows that can produce highly accurate digital models, whilst requiring little expertise and can produce highly accurate 3D digital models using a consumer camera and freely available online software.
The first step is to acquire photographs - the number of photographs that are needed vary according to both the specimen's complexity and to the resolution required of the digital model (from as little as three to several thousand). To produce a 3D co-ordinate, any given point has to be present in at least three photographs from different positions.
The next step is to calculate camera positions to create a 3D object. In my study, I used Agisoft Metashape - due to its ability to reliably produce highly accurate 3D models, whilst being relatively inexpensive ($59 for an academic license), having an easy-to-learn intuitive interface, and not requiring an Nvidia graphics to function. Open-source alternatives exist (e.g. Meshroom, COLMAP), but stringent hardware requirements and often less-streamlined workflows means they are slightly harder to use here.
After the photos are loaded into Metashape, they need to be aligned. During this stage, Metashape searches for common points on photographs and matches them and finds the position of the camera for each picture and refining camera calibration parameters, forming a sparse point cloud and a set of camera positions, which can be inspected and incorrectly positioned photos can be removed. The accuracy of the alignment procedure can be altered - whilst higher accuracy settings help to obtain more accurate position estimates, they also increase processing time. Enabling image pair preselection (a subset of image pairs are selected and matched) helps speed this process up. The upper limit of both key (points on a photograph that the program deems important features) and matching (common points in multiple photos) points in any image can be adjusted on every image.
After this, Metashape calculates the depth information for each image, and multiple reconstruction parameters can be specified (e.g. reconstruction quality - higher-quality settings create more accurate and detailed models, but also have longer processing times). Once this is done, Metashape creates a 3D mesh - and a texture can be built to add colour information.
Comments
Post a Comment