Building Detection and Description

Abstract

We present an approach for detecting and describing complex buildings with flat or complex rooftops by using multiple, overlapping images of the scene. We find 3-D rooftop boundary hypotheses from the line and junction features of the images by applying consecutive grouping procedures. First, 3-D features are generated by grouping image features over multiple images, and rooftop hypotheses are generated by neighborhood searches on those features. Probabilistic reasoning, level-of-details, and cues from image-derived unedited elevation data are used at various stages to manage the huge search space for rooftop boundary hypotheses. Three-D rooftop hypotheses generated by above procedures are verified with evidence collected from the images and the elevation data. Expandable Bayesian networks are used to combine evidence from multiple images. Finally, overlap and rooftop analysis are performed to find the final building models.

Input Images & DEM (Digital Elevation Map)

Image Features

All lines

Filtered Lines (by location and orientation) and Junctions

Multi-View Feature Grouping

All Line Groupings

Filtered Lines (by heights) and Junctions

Grouping near-by Line-groupings for Level-of-Details

Hypothesis Generation

  • Polarities of line/junction groupings (of which side the building lies) are assigned.
  • Chains are generated by grouping neighborhood features (line/junction groupings which share the same 2-D lines).
  • Use parallel relationships to group near-by chains.
  • Find closures (figure).

    Hypothesis Verification & Overlap Analysis

  • Apply EBN's for hypothesis selection, verification, and overlap analysis.
  • Evidence used: positive/negative roof line support (RP, RN), wall vertical lines (WV), darkness of shadow area (SD), and DEM cue coverage (DEM1).
  • Additional information: size of the hypothesis (Size) and projected length of the wall vertical line (WL).

    • Roof Analysis

    Eaves (Boundaries)

    Find Hip Lines

    Find Ridge Lines

    Find Superstructures

    Results



    Publications

  • Z. Kim and R. Nevatia, "Automatic Description of Complex Buildings from Multuiple Images," Computer Vision and Image Understanding, vol. 96, no. 1, pp. 60-95, 2004.
  • Z. Kim, Multi-View 3-D Object Description with Uncertain Reasoning and Machine Learning, Ph.D. Thesis, CS Dept, USC, 2001. [PDF File]
  • Z. Kim, A. Huertas, and R. Nevatia, "Automatic Description of Buildings with Complex Rooftops from Multiple Images", Proc. IEEE Computer Vision and Pattern Recognition, vol 2., pp. 272-279, 2001. [PDF file] [PDF file, Black&White]
  • Z. Kim, A. Huertas, and R. Nevatia, "A Model-Based Approach for Multi-View Complex Building Description," Proc. ASCONA, 2001. [PDF file]
  • Z. Kim, A. Huertas, and R. Nevatia, "Automatic Description of Complex Buildings with Multiple Images," Proc. 5th IEEE Workshop on Applications of Computer Vision, pp. 155-162, 2000. [PDF file]