Graph Matching for Road Network Retrieval
190 likes | 328 Vues
Graph Matching for Road Network Retrieval. Avik BHATTACHARYA. Image Database at ARIANA. ENST/CNES SPOT5 images SupCom, TUNISIA images Images selected with few global scenarios. Few examples of selected images. Continued …. Queries. Queries with relevance to these databases for
Graph Matching for Road Network Retrieval
E N D
Presentation Transcript
Graph Matching for Road Network Retrieval Avik BHATTACHARYA
Image Database at ARIANA • ENST/CNES SPOT5 images • SupCom, TUNISIA images • Images selected with few global scenarios
Queries • Queries with relevance to these databases for • IKONA System • KIM System
Scenarios • Global scenarios selected at ARIANA • Urban/Semi urban/Urban or Semi urban with complex junction/Urban with perpendicular junctions • Rural areas • Airport • Mountain roads
Scenarios from IGN • A meeting with Sylvain Airault at IGN • Pre-knowledge information from cartographer, i.e., importance of the road, actual width of the road or junction • Physical bounds of road structures, i.e., for crossroads and circular junctions
Ideas of graph matching • Probabilistic arguments towards graph matching. • Could reduce the complexity of the matching problem • Metric based clique comparison • Hamming distance, Levenshtein distance, Hausdroff distance, etc.
A survey of road extraction methods • Linear filtering • Mathematical morphology • Variational methods • Markov fields • Neural networks • Dynamic programming • Multiresolution analysis
2 Methods Used For Road Extraction in ARIANA, INRIA, Sophia • Variational Method (By Marie Rochery, present Phd Student at INRIA, Sophia). • Stochastic Method (By Caroline Lacoste, past Phd Student at INRIA, Sophia).
Detector Characteristics • Algorithms with specific goals • Input data, e.g., intensity, edges, lines • Resolution • External knowledge, e.g., GIS, geographical database • Context, e.g., rural/urban roads, linear elements • Output, e.g., pixels, attributes, polygons, segments
The Approach • The node and edge attributes could comprises of : • Spectral properties, e.g., surface characteristics • Geometric properties, e.g., steepness, width, curvature • Topological properties, e.g., road links, networks • Contextual properties, e.g., max width, max curvature
The Approach • Starting from the extracted image (Rochery’s work, Variational method) construct the shock graph. • Shocks Vs Skeleton • “The Shocks form along the reaction axis reduces to traditional skeleton when information regarding type, group, and salience is discarded”
The Approach • The present difficulties • Attempts to locate shocks at grid points suffers from discretization artifacts. • Equivalent graph representation of the road network for both the extraction methods.
Future Work • Node and Edge attributes • Geometrical attributes ! • Junction angles and resulting histogram and network density • The graph construction from Lacoste’s work (Stochastic method)