3D reconstructions for monitoring the archaeological site of Tzintzuntzan, Mexico using Structure-from-Motion and Multi-View-Stereo algorithms

Abstract

Monitoring of archaeological sites is crucial for identifying and addressing structural damage, and this is especially relevant to historically rich sites like Tzintzuntzan in Michoac´an, Mexico, where harsh weather conditions and tourist visitation pose a threat to the pyramid structures. It is imperative to survey the structure and its possible damage frequently, cheaply and at high temporal and spatial resolutions. Recent advancements in Structure from Motion and Multi-View Stereo pipelines paired with UAV imagery make 3D reconstruction through SfM-MVS a fast and cost-effective approach, ensuring data can be captured rapidly and consistently and later used for volumetric and distance-based change detection. This work aimed to study the capabilities of several chosen SfM-MVS software and datasets for the purposes of change detection on the Tzintzuntzan archaeological site, given data limitations such as lack of dedicated ground truth, top-down imagery, and high variability in the parameters used when the available data was gathered. Three datasets, taken across 2015, 2017, and 2020, were analyzed for the purpose of accurate change detection on their 3D point-cloud reconstructions. Particular focus was made on reporting the effects of internal camera parameters and data-gathering parameters such as flight path, altitude, and image angle and overlap on final 3D reconstruction accuracy and consistency. Results indicated that the Pix4Dmapper SfM-MVS software provides optimal reconstruction density and accuracy the quickest. Its fish-eye lens distortion model makes Pix4Dmapper applicable to images taken with a wider range of cameras. Good initial internal parameter inputs for SfM positively affected reconstruction quality, being imperative for datasets with greater flight-path limitations. For the purpose of providing accurate initial parameter inputs, geometric camera calibration is the most straightforward and consistent approach and remains an important factor to consider in image-gathering missions for archaeological sites. Despite the data limitations, SfM-MVS applied to UAV imagery proved a viable method of monitoring the state of the archaeological structure in a cost-effective and flexible way, producing high-quality outputs when specific parameters were taken into account during data capture and the SfM-MVS reconstruction process.