Critique of: Automatic and Accurate Extraction of Road Intersections from Raster Maps by Yao-Yi Chiang · Craig A. Knoblock · Cyrus Shahabi · Ching-Chien.

Critique of:Automatic and Accurate Extraction of Road

Intersections from Raster Mapsby Yao-Yi Chiang · Craig A. Knoblock · Cyrus Shahabi · Ching-Chien Chen

Vikram Reddy Donthi ReddyToufong VangCSCI 8715September 20, 2011

Problem Statement

Difficult to accurately and automatically extract road intersections from raster maps.

Significance of Research Spatial Data set for rasters does not

include or identify road intersections. (Road intersections are object-based models .)

These features may be used to combine or process spatial data sets.

Developed a framework for accurate and automatic separation of specific object-based models from raster.

Difficulty in Accomplishing Maps are complex. Computers have difficulty

distinguishing map features and elements from one another.

Current methods require user input/intervention to process.

Contributions

Major Contributions of the Research Developed method for automatically extracting road data.

95% precision. 75% completeness.

Researchers’ method does not require prior knowledge of the map.

Most significant? 95% accuracy + 75% completeness. Automatic extraction of map data.

Why? Rapid development and integration of

data where none may exist.

Key Concepts

Chiang et al, 2009

Go from raster image of Tehran…(Google map.)

…to hybrid map.(Google map + Tourist map.)

Key ConceptsThe Approach

1. Automatic Segmentation

Raster Map

2. Extract and rebuild road layer

3. Identify road intersections and extract.

Binary Map

Road Layer

Road Intersection Pts, connectivity, and orientation

Key Concepts

Segmentation (remove background).

Researchers premise: foreground colors has high contrast to background colors.

Key ConceptsPreprocessing (extract road layers a and b).

Rebuild road layer (c and d).

ID and extract road data (e).

Validation Methodology

Related Work

Utilized related research and methods.Segmentation process. Road extraction and rebuild.

Researchers’ Prior Work

Localized template matching (LTM)(compare experiment results with original raster)

The Experiment

Validation Methodology

Verification of accuracy of process.

Geometric Similarity(Lay term: how close is the extracted point from to the original point on the raster?)

Evaluation

CritiqueResearch assumption

1) Road lines are straight within small distances.

2) Linear structures are mainly roads.

Falls apart when handling canals and other man-made non-road features.

Revisions?Framework Straightforward. Solid.

Add… Process for handling artificial map features that are not necessarily roads (e.g.,

canals)