You are here

Technical report | Evaluation of Available Software for Reconstruction of a Structure from its Imagery

Abstract 

In this report the Computer Vision System toolbox of MATLAB® and the Visual Structure from Motion software are evaluated on three datasets of airborne imagery, provided by Defence Research and Development Canada, and on a dataset of synthetic imagery generated by the VIRSuite software developed in the Defence Science and Technology Group. The user interface for a developed code for structure reconstruction, based on the feature detection algorithms of the MATLAB toolbox, is described.

Executive Summary

The objectives of the "Three-Dimensional Target Reconstruction" International Collaboration Project (CP 7-25-13) between Australia, Canada and USA, are to extract three-dimensional information from a target via low to high cost vision sensors on moving ground or air platforms through algorithm development. Methods of analysis include comparison of three-dimensional reconstruction algorithms to `truth' data gathered via Laser Detection and Ranging (LADAR) sensors including inertial sensors.

Existing techniques allow reliable reconstruction of smooth surfaces, such as terrains, but the reconstruction of three-dimensional buildings is very challenging with automated software. This project is aimed at investigating new techniques to improve robustness of the reconstruction process, and also looks at the issues of using LADAR data for direct reconstruction. LADAR data is often sparse, particularly at longer ranges, and the fusion of two-dimensional imagery can be beneficial. The expected defence outcomes from the project are:

  • the development of robust techniques for mission planning when using imaging seekers
  • collaboration on algorithmic techniques for automatic target reconstruction, detection and recognition
  • improved simulation techniques and sharing of test data sets.

This report partially addresses the objectives of the collaboration project. The Computer Vision System toolbox of MATLAB® and the Visual Structure from Motion (VisualSFM) software are evaluated on three datasets of airborne imagery, provided by Defence Research and Development Canada, and on a dataset of synthetic imagery generated by the VIRSuite software developed in the Defence Science and Technology Group. The user interface for a developed code for structure reconstruction, based on the feature detection algorithms of the MATLAB toolbox, is described.

Key information

Author

Leonid K Antanovskii

Publication number

DST-Group-TR-3356

Publication type

Technical report

Publish Date

April 2017

Classification

Unclassified - public release

Keywords

Science, Mathematics, Algorithms, Computer Vision, Structure Reconstruction