From 2D images to a 3D model

The creation process of this 3D model can be divided into three main tasks:

• Capturing photographs;
• Pre-processing the images;
• Developing the model.

Task 1: Capturing photographs

The figurine was photographed at the Cyprus Museum of Natural History, where it is currently stored. The photographs were taken using a Canon EOS 850D camera with the following settings: f/8 aperture, ISO 200, 1/15 sec exposure time, and an 18 mm focal length. A tripod and remote camera control were used to prevent blurriness and ensure image sharpness. I did not have a lightbox where to put the figurine to guarantee constant light during the entire photoshoot cycle and isolate the object, but I set a white background using paper sheets and cardboard in order to prevent an extremely noisy background. In photogrammetry, software algorithms identify common points (tie points) in overlapping areas of images, allowing them to merge the photographs accurately (fig. 1). To ensure complete coverage of the object, the figurine was, therefore, photographed from multiple angles and directions.

Fig. 1 - Result of the alignment process where camera
positions show that the  complete coverage
of the subject was reached

Task 2: Pre-processing the images

Having a high-quality set of images is essential for achieving a well-detailed, photogrammetry-based 3D model. Due to lighting issues at the photography setup, pre-processing was necessary to correct the lighting of the images. The RAW files were adjusted and enhanced accordingly.

Task 3: Developing the model

The model was created using the Structure from Motion (SfM) technique, a photogrammetric method that combines the accuracy of photogrammetry with the automation and detailed object recognition provided by computer vision.

The software used for processing the data was Agisoft Metashape 2.2.0 Professional Edition. The workflow for the 3D model creation followed these main steps: 

• Loading images into the software
• Assessing image quality and disabling images with a quality value lower than 0.5
• Manually creating image masks
• Aligning photos (fig. 1)
• Building the point cloud
• Building the mesh (3D polygonal model and block model)
• Building texture
• Refining the model, primarily by closing gaps 

The model was then scaled to ensure an accurate representation of the object’s actual dimensions.
While the final model had high geometric quality, minor issues arose with the final part of the dress figurine and the base. The object currently stands with the support of a metal rod inserted between the body and a marble base, which were added during its inventory process to keep the figurine upright. However, due to the reflective properties of the metal, the rod could not be accurately reconstructed in Metashape. In addition, the reflection caused by both the rod and the marble contributed to creating some noise between the dress bottom and the base (fig. 2). To solve this, I opted to mask the rod and generate the final model without it. In a further step, I imported the textured model into Blender, a free and open-source 3D creation suite, where the metal rod was manually recreated using the “cylinder shape add” function (figg. 3, 4). 

Fig. 2 – First attempt of building texture with noise
between the dress and the base of the figurine

                  Fig. 3  – Final figurine model without rod                                 
  
Fig. 4  – Building rod process in Blender