The project is spearheaded by DHI InnoTech in collaboration with Royal Bhutan Police.
Project description
The ANPR system addresses a critical gap in Bhutan’s public safety infrastructure. Existing CCTV surveillance networks, including the Safe City Project cameras deployed across Thimphu, Gelephu, and Sarpang, rely entirely on manual human monitoring with no automated vehicle identification capability. This creates unacceptable delays in crime detection, makes ecological zone vehicle limits unenforceable, and leaves border security dependent on manual checkpoints.
Foreign ANPR solutions — predominantly sourced from Israeli and other international vendors — are prohibitively expensive for Bhutan’s scale, require extended restoration times when systems fail (replacement parts and vendor support must be sourced from overseas), offer no timely in-country maintenance (foreign vendors lack local presence, with no guaranteed response times), and run on legacy architectures designed for standardised international plates that cannot handle Bhutan’s unique plate formats (BG, BP, BT categories) or Dzongkha script.
DHI InnoTech developed a fully sovereign alternative. The system employs a multi-stage AI inference pipeline: YOLOv8 for vehicle detection and number plate localisation, followed by a custom OCR engine built on Microsoft Florence 2, fine-tuned specifically on thousands of Bhutanese number plate images collected from real traffic across Thimphu, Gelephu, and Sarpang. A post-processing layer applies Bhutanese plate format validation rules and character-level confidence correction. The production system uses HikVision IDSTCM403BI cameras with IR supplementary lighting capable of capturing plates at vehicle speeds up to 120 km/h, processed by NVIDIA RTX 3080 GPU servers connected via MQTT for real-time stream ingestion. The web application provides a dashboard with search, blacklisting, analytics, and report export functionality.
What We Are Seeking:
We are seeking technology and research partners who can strengthen our AI model development capabilities, particularly in multilingual OCR and edge deployment optimisation. Specifically, we are looking for collaboration with CVUT (Czech Technical University in Prague) to co-develop advanced computer vision architectures for number plate recognition, provide structured training for our engineering team in vision transformer models and inference optimisation (TensorRT, ONNX quantisation), and jointly build a region-adaptable ANPR product for South Asian markets.
Why this Matters Now
Bhutan is at a pivotal moment in its national security and smart city infrastructure development. The Gelephu Mindfulness City initiative is driving unprecedented investment in modern urban systems, and the Safe City Project is actively expanding camera coverage across the country. Without an automated vehicle identification layer, this investment in physical surveillance infrastructure remains fundamentally underutilised — cameras capture footage, but no system can automatically process it.
Why Bhutan and Why InnoTech
DHI InnoTech’s advantage lies in its position as Bhutan’s only domestic AI development capability with direct institutional ties to the national holding company (DHI) and an established working relationship with the Royal Bhutan Police. Unlike foreign vendors, the DRIVE AI team operates in-country, understands Bhutanese regulatory and operational context (BCTA, eRaLIS, mBOB), and can provide same-day maintenance and support. The team has already collected, annotated, and trained models on real Bhutanese traffic data — a sovereign dataset that no foreign vendor possesses.
Investment Required/Project Cost:
Phase I (Completed): Nu. 2,966,882 — covering 10 AI-enabled ANPR cameras, 3 GPU servers, supplementary lighting, software development, installation, and one-year post-handover maintenance across Gelephu (4 cameras), Sarpang (4 cameras), and Thimphu (2 cameras).
Phase II (Planned): Estimated Nu. 5–7 million for expansion to 20–30 additional camera sites across Bhutan’s national road network, including highway checkpoints, additional urban intersections, and border crossings. Includes eRaLIS database integration, speed detection module development, and mobile ANPR unit prototyping.
Partnerships and Collaboration
Royal Bhutan Police (RBP): Primary client and operational partner. RBP provides site access, security clearances, operational requirements, and end-user adoption. The Phase I contract (signed 18 April 2025) includes provisions for future expansion under the broader Safe City initiative.
Want to collaborate on this project? Write the lead directly, or pitch an adjacent idea through the open queue.
We are also seeking funding and institutional support to scale deployment across Bhutan’s national road network (Phase II and beyond), covering additional Safe City sites, highway checkpoints, and border crossings. Investment is required for expanded hardware procurement (cameras, GPU servers, edge devices) and software feature development including integration with Bhutan’s national vehicle registration database (eRaLIS).
Simultaneously, Bhutan faces growing cross-border security challenges, increasing vehicle volumes on its road network, and the need to enforce ecological zone regulations in environmentally sensitive areas. The Phase I deployment has proven the technical viability and operational value of the system. There is now a narrow window to scale deployment before the Safe City infrastructure is locked into configurations that exclude AI integration, and before competing vendors establish footholds with expensive, hard-to-maintain foreign systems.
The intellectual property developed under the ANPR project remains the exclusive property of DHI, giving Bhutan full control over its vehicle surveillance technology without dependency on foreign licence agreements, recurring subscription fees, or external vendor goodwill. This sovereignty extends to the AI models, software code, system architecture, and training datasets.
Annual Maintenance Contract (AMC): 10% of overall project cost per year (applicable from Year 2 onwards), covering software updates, hardware repairs, firmware updates, troubleshooting, and system health monitoring
CVUT (Czech Technical University in Prague): Sought as a research and technology partner for joint development of advanced OCR architectures, knowledge transfer in vision transformers and edge deployment optimisation, co-development of a region-adaptable ANPR product for South Asian markets, and access to standardised benchmarking frameworks.
DANTAK (Border Roads Organisation): Prospective domestic client. DANTAK manages strategic road infrastructure in Bhutan where automated vehicle monitoring would significantly enhance security and traffic management along key routes.
Regional Market Expansion: The Florence 2-based OCR pipeline is designed to be re-trainable on regional plate datasets. Active exploration of go-to-market strategies for India and Nepal, where similar challenges exist around non-standardised plates, multilingual scripts, and the need for sovereign, locally-hosted recognition systems.
The Bhutan Sovereign LLM Initiative, led by DHI InnoTech, aims to build a national AI layer for Bhutan, which is a secure, sovereign Large Language Model integrated with a structured National Data Lake. This platform will:
Aggregate structured and unstructured national datasets.
Serve as a sovereign knowledge engine trained on Bhutan-specific data.
Power AI agents capable of policy analysis, institutional intelligence, and public-facing services.
Act as a digital extension of Bhutan. It will be a responsible AI representation of the Kingdom for both domestic governance and international engagement.
The initiative envisions Bhutan not merely as a consumer of global AI systems but as a creator of a nationally aligned intelligence layer, grounded in Bhutanese values, governance structures, and development priorities.
The deployment architecture remains flexible:
Phase 1 (Prototype): Agent-based architecture built on leading AI APIs.
Phase 2 (Conditional on Funding): Migration to sovereign GPU infrastructure hosted within Bhutan.
Alternative Path: Continued operation on secure global cloud infrastructure if capital expenditure is constrained. The system is being developed as modular, interoperable, and sovereign by design.
Our objective is to establish a national-scale, technology-enabled GLOF risk monitoring and mitigation platform, beginning with high-risk catchments such as the Punatsangchhu basin.
RiKhaNET addresses a core operational challenge in Bhutan’s tiger landscapes where field data and incident information are often fragmented across tools and stakeholders, limiting timely prevention. RiKhaNET consolidates these inputs into a single, governed platform by integrating Bhutan’s SMART (Spatial Monitoring and Reporting Tool) data with AI-driven analytics and real-time sensor inputs. Using machine-learning models, the system generates dynamic risk heatmaps, geofencing alerts, and automated notifications to help rangers and communities anticipate conflict hotspots before incidents occur, shifting response from reactive to proactive prevention.
The Stakeholder meetings conducted emphasized that the pilot must be designed to strengthen trust and adoption, not only technology performance. Three design requirements were highlighted:
Data governance and transparency: The current hosting arrangements raise concerns about filtering, access control, and sensitive wildlife-location data; governance must be explicit and communicated clearly.
Community safety communication: The alert systems must be carefully designed so alerts support prevention and coordinated action rather than increasing fear; community focal persons can strengthen trusted communication.
Operational realism for detection: It has been noted that tiger detections can be rare; deployments may run for months yet capture limited tiger events, so the pilot should prioritize focused hotspot monitoring with sufficient duration to produce meaningful training and pattern data.
An IoT-enabled Smart Water Management System designed to improve water distribution, reduce wastage, and ensure reliable access to safe drinking water. The system integrates smart sensors for water quality, pressure, and flow monitoring, supported by remote monitoring and control to detect anomalies and optimize water supply from source to end users. Currently being piloted in Bhutan, the solution has potential for regional and international scale-up.
