acid vision

Improved methods for automatically reconstructing 3D digital models from multiple digital photos.

overview

The ACID Vision project explored methods and tools for creating accurate 3D models from digital images very quickly and with little or no user interaction. A technology demonstrator was made available through the free 3Dsee web service where photos could be uploaded and bump maps (used as complex textures in 3D modeling) and 3D models derived from them could be viewed and downloaded.

research aims

The aim of the ACID Vision project was to research and develop:

• A robust and accurate feature extraction and tracking module
• Robust plane based calibration (for bump mapping) able to detect degenerate planar surfaces
• Robust turntable based calibration (for high-definition modelling)
• Free standing 3D object reconstruction
• Plane-based 3D bump-map reconstruction
• Effective user interfaces and workflows for bump-map and free-standing object modeling

research outcomes

The ACID Vision project incorporated best-practice research in 3D reconstruction and overcame a range of practical and technical issues to create a functional and robust system for 3D reconstruction. A number of technical innovations were developed to support robustness when dealing with real cameras and images, including new theory for underlying components of the reconstruction process and novel interfaces to enable users to interact intuitively with the system.

Research into 3D reconstruction considers the formation of light (from the real world) as it appears on the lens of a camera, and what the content within these images allows a system (set of algorithms) to determine about the precise nature of that environment. 3D reconstruction research comprises several areas of study including feature extraction/matching and segmentation, multiple-view geometry, and dense 3D reconstruction.