3D Pose Estimation Oriented to the Initialization of an Augmented Reality System Applied to Cultural Heritage

Abstract

Augmented reality (AR) applied to cultural heritage intends to improve the learning experience in archaeological sites, not only for visitants but also for researchers. 3D Pose estimation is a common problem in applications for AR, object recognition, 3D modeling, among others. AR systems use different methods to estimate the camera pose: edge detection and key-point detection among others. The choice of the method to be used depends on the features of the scenario to be detected. In this work, a comparison study of the main 3D model-based pose estimation methods is performed. In addition, we present the implementation and validation of a pose estimation algorithm, oriented to the initialization of an AR system applied to “Huaca de la Luna”, an adobe brick pyramid built by the Moche civilization in the northern Peru. The proposed algorithm presents two phases, a training phase, where 3D key-points are extracted from a reference image, and a detection phase, where the initialization process is performed by comparing 2D/3D points correspondence using a PnP algorithm. We have compared four variations of the 3D pose estimation algorithm using different methods: SIFT and SURF descriptors for key-point description and EPnP and REPPnP algorithms for PnP pose estimation. Results show a translation error of 1.54 cm, with a mean processing time of 2.78 s, a maximum re-projection error of 1.5 pixels and a successful estimation rate of 100% in scenarios with normal and high light conditions.