A Neural Field-Based Approach for View Computation & Data Exploration in 3D Urban Environments

Area of Science:

• Computer Vision
• Urban Informatics
• Geographic Information Systems

Background:

• 3D urban datasets are increasingly available but difficult to analyze due to computational limits and complex geometry.
• Occlusion and manual viewpoint adjustments hinder efficient exploration of large-scale 3D urban environments.

Purpose of the Study:

• To develop a novel view-based approach for efficient 3D urban data exploration.
• To introduce a neural field-based method for implicit representation of 3D environments.
• To enable faster data queries and facilitate the search for optimal viewpoints, addressing occlusion problems.

Main Methods:

• A view-based approach using a vector field to encode environmental views.
• A neural field-based method for creating efficient implicit representations of 3D urban data.
• Direct queries for view assessment and inverse queries for occlusion avoidance and pattern matching.

Main Results:

• The proposed method enables faster direct and inverse queries on 3D urban data.
• Effectively identifies desirable viewpoints and analyzes visibility for urban planning tasks.
• Demonstrated success in visibility assessments, solar exposure evaluation, and visual impact analysis.

Conclusions:

• The neural field-based approach significantly enhances the efficiency and effectiveness of 3D urban data exploration.
• This method supports critical urban analysis tasks, improving decision-making in urban planning and development.
• Publicly available code and data facilitate broader adoption and further research in the field.