This project was part of a Diploma in Industrial Studies dissertation on Leveraging Temporal Coherence to Streamline 3D reconstruction, written in 2022.
The general idea was to quantitatively assess the impact of exploiting temporal coherence among video frames, when utilised as an ordered input dataset, for the purpose of reconstructing a partial three-dimensional model from two-dimensional images. In other words: shoot a video and produce a partial 3D mesh from it.
This research entailed a stage-wise temporal analysis of various feature matching techniques using standard COLMAP methods, encompassing exhaustive, sequential, and transitive matching, among others.
The main objective was to streamline the procedures involved in 3D reconstruction for generating partial meshes by eliminating redundant steps typically required when dealing with unordered datasets.
Github repo: https://github.com/raunak-h/v2o