paper · 2026

BSI as a Coordinate-Free Linear Reference System

The Korean Road Name Address Act mandates a monotonically increasing integer along every named road. That sequence is formally an ISO 19148 Linear Reference System measure — but has never been used as one. Activating it gives O(1) integer-arithmetic distance, a Dual Graph that scales over 100× better under routing constraints, and 24 ms district-level routing on commodity hardware.

What this paper does

The Korean Road Name Address (RNADDR) system assigns every addressed point a road name and a Basic Section Index (BSI) — an odd integer enumerating building plots from the road origin at fixed 10 m (gil) or 20 m (ro/daero) intervals. The sequence is statutorily maintained, monotonically increasing, and distance-preserving.

It is — formally — a Linear Reference System measure that satisfies the ISO 19148 axioms. But every existing LRS implementation (ArcGIS Roads & Highways, Norway NVDB, US ARNOLD) derives its measure from coordinate geometry. None has ever used a civic address number as the measure itself.

This paper activates BSI as the LRS measure directly. The result is a coordinate-free spatial framework: routing, distance computation, and attribute queries all run on integer arithmetic with no GPS coordinates, no spatial database, and no geocoding at any step.

Concrete artifacts

Distance formula. d(A, B) = |b_A − b_B| / 2 × δ_r — O(1) integer arithmetic. Validated against WKT polyline ground truth at 2.1 % mean error for paths over 500 m.

Two intersection tables, built from 369,597 national road records. The Road Name Address Act classifies roads by width only and does not specify any topological relationship between tiers. We construct the implicit topology:

  • intrsct_index — 205,433 rows recording where each gil meets its parent ro/daero, across all 258 city/county units.
  • road_conn — 79,760 rows recording arterial-to-arterial junctions, stored bidirectionally.

Dual Graph. Named roads as nodes, intersections as edges. Adding routing constraints (turn restrictions, time-of-day rules, vehicle classes) doesn’t grow the node count. In the conventional Primal Graph the count grows multiplicatively — 3.5× unconstrained, over 100× under combined constraints.

Working implementation. ~113 MB in-memory SQLite. Gangnam-gu mean routing latency: 24 ms.

Why it matters

Korea’s existing ITS infrastructure represents roads as Node/Link segments — one identifier per intersection-to-intersection span, one aggregated attribute per span. Three deficiencies follow:

  • Resolution loss. A 1 km link carries one average speed value. Congestion in the final 200 m looks like the whole link is congested.
  • Directional blindness. A single link value cannot distinguish a free-flowing through-lane from a 500 m right-turn queue at the same intersection.
  • Discontinuity. Tracking one road end-to-end requires joining multiple link IDs at every intersection.

BSI binds attributes at 10–20 m granularity, supports per-direction binding, and assigns one ID to an entire road. Critically, this is not a replacement: existing ITS link IDs map to BSI ranges via a one-time conversion table, so adoption is incremental.

Status

First in a three-paper series. Korean patent application No. 10-2026-0039486 (filed 2026-03-05). PCT international filing planned within the priority year.

The author is actively seeking collaborators in autonomous driving infrastructure, ITS modernisation, public transit, indoor address standardisation, traffic congestion modelling, and smart-city data standardisation.