Seoul Smart City Technology Stack — IoT, Digital Twin, 5G, and AI Infrastructure Explained

Architecture Overview — Seven Layers of Urban Intelligence

Seoul’s smart-city technology stack is not a single system but a layered architecture of interconnected platforms, each owned by different city agencies but sharing data through standardized APIs and a centralized data lake. The stack processes 48 terabytes of urban data daily — from environmental sensors, traffic cameras, transit systems, energy meters, and citizen interactions — and routes that data through analytics pipelines that inform real-time operational decisions and long-term urban-planning models.

The architecture can be decomposed into seven functional layers:

This guide examines each layer in technical detail, with cross-references to dedicated pages covering individual systems.

Layer 1: Sensing — 85,000 Data Collection Points

Seoul’s sensing infrastructure encompasses 85,000 discrete data-collection points across multiple sensor networks:

S-DoT Environmental Sensors

The S-DoT sensor network deploys 1,100 multi-parameter environmental sensors (scaling to 50,000 by 2030) that capture 17 atmospheric, acoustic, and optical data types at two-minute intervals. Each S-DoT unit transmits via LoRaWAN or LTE-M to the Seoul Data Center, where raw telemetry enters the Kafka-based ingestion pipeline. S-DoT data feeds air-quality dashboards, noise-complaint heatmaps, urban-heat-island models, and the S-Map digital twin.

The 812 integrated smart poles extend S-DoT sensing with traffic cameras, WiFi access points, emergency buttons, and EV charging — consolidating five to seven pieces of street furniture into a single mast.

Intelligent CCTV Network

Seoul operates over 83,000 CCTV cameras across its 25 districts, of which approximately 15,000 are equipped with AI-powered video analytics. The AI CCTV system performs real-time detection of 14 event categories: traffic accidents, illegal parking, abandoned objects, crowd density exceeding safety thresholds, fire/smoke, pedestrian falls, vehicle wrong-way driving, and aggressive behavior. All video analytics run on edge-compute units (NVIDIA Jetson or equivalent) co-located with camera clusters, with only metadata — not raw video — transmitted to central servers.

The edge-processing architecture serves two purposes: it reduces bandwidth requirements (metadata at 2–5 Kbps per camera vs. raw video at 4–8 Mbps), and it addresses privacy concerns by ensuring facial imagery never leaves the camera location. The Personal Information Protection Act (PIPA) mandates that CCTV footage containing identifiable faces be stored for no more than 30 days and accessed only through audited, role-based-access-controlled interfaces.

Traffic Detection Infrastructure

The TOPIS transport management system receives data from 5,200 inductive-loop detectors embedded in road surfaces, 3,400 overhead vehicle-detection cameras, 1,200 Bluetooth travel-time sensors, and GPS telemetry from 8,200 public buses and 72,000 registered taxis. This multi-source traffic data provides real-time vehicle counts, speeds, and travel times across 98 percent of Seoul’s arterial road network.

Utility and Infrastructure Sensors

Beyond the visible sensing infrastructure, Seoul operates embedded sensor networks in critical infrastructure:

• Water-supply network: 4,200 pressure and flow sensors monitoring 13,700 km of water mains, feeding a leak-detection AI that reduced water loss from 8.2 percent to 4.1 percent between 2018 and 2025
• Gas network: 2,800 methane-detection sensors along the 17,400 km gas distribution network
• Sewage system: 1,600 level and flow sensors in storm drains and combined sewer systems, integrated with the flood-early-warning system
• Bridge structural monitoring: 340 accelerometers, strain gauges, and displacement sensors on the 31 bridges spanning the Han River and 420 overpass structures

Layer 2: Connectivity — The 5G and Fiber Backbone

5G Infrastructure

Seoul’s 5G network achieves 93 percent population coverage through 148,000 base stations operated by SK Telecom, KT, and LG U+. The network operates on two frequency bands: 3.5 GHz mid-band (providing broad coverage at 500–900 Mbps) and 28 GHz mmWave (providing localized ultra-high-speed coverage at 2–4 Gbps in dense commercial areas). For smart-city applications, 5G provides the low-latency connectivity required by time-critical systems — autonomous-vehicle infrastructure-to-vehicle communication, real-time CCTV streaming, and emergency-alert distribution — that cannot tolerate the 50–100 ms latency typical of 4G LTE.

SK Telecom’s dedicated 5G network slice for Seoul Metropolitan Government applications guarantees sub-10 ms latency and 99.99 percent uptime for priority traffic including TOPIS data feeds, emergency-dispatch communications, and S-DoT telemetry backhaul.

LoRaWAN for Low-Power IoT

For sensor networks that prioritize battery life over bandwidth — S-DoT environmental sensors, water-meter telemetry, and smart-waste bin fill-level monitoring — Seoul operates a LoRaWAN network of 420 gateways providing coverage across all 25 districts. LoRaWAN’s 0.3–50 Kbps data rates are sufficient for periodic telemetry (readings every 1–15 minutes), while its 10+ km range and sub-milliwatt sleep-mode power consumption enable battery lives of 5–10 years for simple sensor nodes.

Fiber Backbone

Seoul’s metropolitan fiber network, operated by KT and managed under contract by the Seoul Metropolitan Government, provides 10 Gbps backbone connectivity between all city agency data centers, the TOPIS traffic-management center, all 25 district offices, and 480 critical-infrastructure facilities. The network uses dedicated dark-fiber pairs separate from commercial internet traffic, ensuring bandwidth availability during peak commercial usage and providing physical-layer isolation for security-sensitive government communications.

Layer 3: Ingestion — The Data Pipeline

Architecture

The Seoul Data Center ingests data from all sensing and connectivity layers through a centralized streaming platform:

1. Message broker: Apache Kafka cluster (24 brokers, partitioned by data source and district) handles real-time message ingestion at peak throughput of 3.2 million events per second
2. Stream processing: Apache Flink consumers perform real-time quality assurance, unit conversion, geospatial tagging, and anomaly detection
3. Time-series store: InfluxDB cluster (12 nodes) serves real-time dashboards and the public Open Data API with sub-second query latency for recent data
4. Analytical store: Apache Parquet files on MinIO object storage feed batch-analytics workloads at the Seoul Big Data Campus
5. Data catalog: Apache Atlas metadata management tracks data lineage from sensor to dashboard, enabling auditable data governance

Quality Assurance Pipeline

Every incoming data point passes through a three-stage QA pipeline:

Approximately 0.3 percent of incoming data points are quarantined daily, primarily due to sensor drift in PM2.5 optical counters and GPS jitter in bus telemetry. Quarantined data is reviewed by automated recalibration algorithms and, when necessary, by human operators.

Layer 4: Analytics — Machine Learning at Urban Scale

Traffic Prediction and Optimization

The AI traffic management system processes real-time traffic data from all 5,200 inductive-loop detectors and 3,400 cameras to predict traffic conditions 15, 30, and 60 minutes into the future across Seoul’s 6,100 signalized intersections. The prediction model — a graph neural network (GNN) that treats intersections as nodes and road segments as edges — achieves 92 percent accuracy at the 15-minute horizon and 84 percent at the 60-minute horizon.

Predictions feed directly into the adaptive signal-control system, which adjusts green-phase timing at 2,800 of Seoul’s 6,100 signalized intersections (the remaining 3,300 are being retrofitted for adaptive control by 2028). The system reduces average travel times by an estimated 15.6 percent on equipped corridors compared to fixed-timing plans.

Environmental Forecasting

S-DoT data feeds a hyperlocal air-quality forecast model that predicts PM2.5 concentrations at 500-meter resolution for the next 24 hours. The model combines S-DoT sensor readings with meteorological data (from the Korea Meteorological Administration), traffic-volume data (from TOPIS), and emissions inventory data (from the Ministry of Environment) using an ensemble of gradient-boosted decision trees and convolutional neural networks. Forecast outputs are published through the Seoul Air Quality API and integrated into navigation apps that offer low-pollution routing.

Predictive Maintenance

Infrastructure sensor data from bridges, water mains, and road surfaces feeds predictive-maintenance models that estimate time-to-failure for critical assets. The bridge-monitoring system, for example, correlates accelerometer data with structural finite-element models to detect fatigue-crack initiation and growth, triggering maintenance alerts 6–18 months before visual inspection would identify the defect. The water-network AI analyzes pressure-wave patterns to detect and localize leaks with 50-meter accuracy, enabling targeted excavation rather than the traditional approach of digging up entire blocks.

Public Safety Analytics

The AI CCTV system processes video from 15,000 cameras simultaneously, generating approximately 2,800 alerts per day across the 14 event categories. Alert accuracy (as measured by the ratio of true positives to total alerts reviewed by human operators) stands at 87 percent for traffic accidents, 78 percent for crowd-density threshold exceedance, and 62 percent for aggressive-behavior detection. False-positive rates are highest for the aggressive-behavior category, reflecting the inherent difficulty of distinguishing threatening from benign physical interactions through video analytics alone.

Layer 5: Visualization — S-Map and Operational Dashboards

S-Map Digital Twin

The S-Map digital twin is a 1:1 three-dimensional model of Seoul that integrates real-time data from all sensing layers into a navigable 3D visualization. Built on a foundation of high-resolution LiDAR scans (5 cm accuracy), aerial photogrammetry, and building-information models (BIM) for 12,000 structures, S-Map represents Seoul’s 605.23 km² of urban terrain as a dynamic, data-rich environment that city planners and agency operators can query interactively.

S-Map supports multiple visualization modes:

• Environmental overlay: Real-time PM2.5, noise, temperature, and humidity from S-DoT sensors, rendered as color-coded gradient layers
• Traffic layer: Live vehicle density and speed data from TOPIS, animated on the 3D road network
• Infrastructure status: Bridge health indicators, water-pressure anomalies, and energy-consumption data for city-owned buildings
• Simulation mode: What-if scenarios for proposed developments, enabling planners to model the traffic, air-quality, and line-of-sight impacts of new buildings before construction permits are issued

TOPIS Operations Center

The TOPIS center — Seoul’s transport-operations nerve center, located in a purpose-built facility in City Hall — displays real-time traffic data on a 22-meter-wide LED video wall and staffs 120 operators 24/7. TOPIS processes 28 million data points per hour from buses, taxis, subway systems, and road sensors, providing the operational interface through which city managers respond to traffic incidents, severe weather, and special events.

Open Data Plaza

Seoul Open Data Plaza publishes 4,700+ public datasets, including real-time feeds from S-DoT, TOPIS, and the subway system. The API platform serves approximately 340 million requests per month from 12,000 registered developers and 800 registered apps. Third-party applications built on Open Data Plaza include hyperlocal weather apps, cycling-route optimizers, real-time bus-arrival displays, and citizen-science air-quality monitoring projects.

Layer 6: Actuation — Closing the Loop

The smart-city stack’s value is ultimately measured not by data collected or models built, but by actions taken. Seoul’s actuation layer includes:

• Adaptive traffic signals: 2,800 intersections with real-time phase-duration adjustment based on AI traffic predictions
• Dynamic bus rerouting: Automatic rerouting recommendations for 360 bus routes during severe congestion, flooding, or air-quality events, communicated to drivers through onboard terminals
• Emergency alert distribution: Multi-channel alert system (cell broadcast, TOPIS displays, smart-pole speakers, app push notifications) capable of reaching 98 percent of Seoul’s population within 90 seconds
• Automated parking guidance: Smart parking systems with real-time occupancy data from 48,000 sensor-equipped spaces, feeding navigation apps that guide drivers to available spots
• Smart street lighting: 165,000 LED streetlights with dimming schedules adjusted dynamically based on S-DoT illumination sensor data, pedestrian-presence detection, and time-of-day policies

The actuation layer executes approximately 14,000 automated actions per day — mostly traffic-signal adjustments and bus-rerouting recommendations — with human-in-the-loop confirmation required only for actions affecting safety-critical systems (emergency alerts, bridge-closure decisions, gas-supply shutoffs).

Layer 7: Governance — Privacy, Transparency, and Citizen Feedback

Data Governance Framework

Seoul’s smart-city data governance follows the “collect once, use many times, share openly” principle, with three tiers of data classification:

Privacy Architecture

The Personal Information Protection Commission (PIPC) provides regulatory oversight of all smart-city systems processing personal data. Seoul’s technical privacy architecture implements:

• Edge processing for all video analytics — raw video never leaves camera sites
• Differential privacy for published datasets containing aggregated personal information
• Data minimization — sensors collect only the data types required for their designated purpose
• Retention limits — CCTV metadata retained for 30 days, environmental data retained indefinitely (non-personal)
• Anonymization pipeline — taxi and bus GPS data published through Open Data Plaza is anonymized using trajectory perturbation algorithms

Citizen Participation

The 120 Dasan Call Center — Seoul’s government helpline — integrates with the smart-city stack to route citizen complaints to the appropriate operational system. A noise complaint, for example, triggers a lookup of S-DoT noise data for the caller’s location, attaches the sensor readings to the complaint ticket, and routes it to the district environmental-management office with both the citizen’s qualitative description and quantitative noise measurements. This data-augmented complaint process reduced average resolution time from 14 days to 6 days between 2020 and 2025.

Integration Standards and Interoperability

Seoul’s smart-city systems communicate through a set of standardized APIs defined by the Smart City Data Hub specification, which the SMG developed in partnership with the Ministry of Land, Infrastructure and Transport (MOLIT) and has published as an open standard for adoption by other Korean cities. The specification defines:

• Data models for 42 urban-data domains (traffic, environment, energy, water, waste, public safety, etc.), based on FIWARE Smart Data Models with Korean extensions
• API protocols using OGC SensorThings API for IoT data and REST/JSON for application data
• Authentication through government-issued API keys with OAuth 2.0 for registered developers
• Data quality metadata including sensor-calibration dates, measurement uncertainty bounds, and data-lineage provenance

This standardization enables new systems to integrate with the existing stack without custom middleware — a lesson learned from the pre-2018 era when each city agency operated proprietary data silos that required expensive point-to-point integrations to share information.

Comparative Benchmarks

Seoul’s advantage is breadth of integration. While Singapore may deploy more cameras and Barcelona more environmental sensors per square kilometer, Seoul’s layered architecture — connecting sensing, analytics, visualization, and actuation through standardized APIs — creates cross-domain intelligence that siloed deployments cannot replicate. The ability to correlate air-quality data with traffic patterns and route bus services accordingly, or to augment video analytics with acoustic data from environmental sensors, produces operational capabilities that exceed the sum of the individual systems.

The Seoul vs. Tokyo smart-city comparison provides additional context on how Seoul’s integrated approach differs from Tokyo’s more decentralized model.