Spatial Data Compression via Linear Topology

Spatial regions defined over a road network—school zones, speed-limit segments, no parking intervals—are conventionally stored as coordinate polygons, requiring spatial joins and floating-point geometry at every query. This paper shows that linear topology admits a fundamental compression: under the Basic Section Index (bsi) linear reference system [3], any such region reduces to a set of integer pairs (na,nb), one per covered road segment, and all spatial queries become integer comparisons. The bsi system thereby realises the conceptual model of ISO 19148 [4] as its most lightweight possible implementation: integer arithmetic on address numbers, with no coordinate transformation at any stage.

Any entity in the Korean Road Name Address (RNADDR)system— pharmacies, hospitals, CCTV installations, school zones—carries a road name and building number that is
already a bsi position, with no geocoding required. This paper formalises a direct attribute binding model built on this property: two attribute types cover the full space of address based road data. Point attributes (infrastructure objects, points of interest) bind as (r,n,v) triples; interval attributes (zones, segments) bind as (r,[na,nb],v) triples. Both are queried by integer arithmetic alone, with no spatial database and no floating-point computation.

A key application is bsi-native facility search: pharmacies, hospitals, or any categorised point of interest are retrieved by integer range scan in O(logn + k), with no coordinate projection. Integrated with the Dual Graph routing of [2],this enables facility-aware routing —nearest pharmacy on path, school-zone speed adjustment—without spatial computation at any stage.

Analysis of a real RNADDR dataset (Seoul Gangseo-gu school zones, 88 records) demon strates that point-type records bind immediately from their road name addresses, while revealing that current zonal data formats publish only the facility point rather than zone boundaries—a limitation of data convention, not of the model