Incremental acquisition of a three-dimensional scene model from images

Area of Science:

• Computer Vision
• 3D Reconstruction
• Robotics

Background:

• Acquiring 3-D models of complex urban environments from images is challenging due to partial information in single views.
• Varying viewpoints and lighting conditions across images necessitate an incremental approach to model construction.
• Existing methods struggle with the complexity and scale of urban scene interpretation.

Purpose of the Study:

• To present the current state of the 3-D Mosaic project for incremental 3-D urban scene modeling.
• To develop a method for refining 3-D models by sequentially analyzing multiple images.
• To integrate stereo analysis, junction-based matching, and a knowledge-based reasoning system for robust reconstruction.

Main Methods:

• Sequential analysis of multiple images to incrementally construct a 3-D model.
• Stereo analysis for extracting 3-D shape information.
• Junction-based matching for sparse wire-frame descriptions in urban scenes.
• A knowledge-based reasoning system to generate approximate scene models from stereo output.
• Acquisition of grayscale information for model faces.

Main Results:

• Demonstrated the utility of junction-based matching for sparse wire-frame extraction in urban scenes.
• Successfully generated approximate 3-D models using a reasoning system combined with stereo data.
• Showcased an experiment combining two views to obtain a refined model.

Conclusions:

• Incremental acquisition from multiple images effectively refines 3-D urban scene models.
• The integration of stereo analysis, junction-based matching, and reasoning provides a robust framework for 3-D reconstruction.
• The 3-D Mosaic project demonstrates a viable approach for building detailed 3-D urban models.