São Paulo Tackles the Urban Duplicate Image Problem — and It's Ahead of Some Major Rivals

São Paulo's city hall quietly crossed a threshold earlier this year: the volume of digital images stored across municipal systems — street-level photography, building permit documentation, flooding-zone surveys, transit camera feeds — surpassed 40 million files. A significant chunk of those, according to a review of procurement documents published by the Secretaria Municipal de Gestão, are duplicates. The problem is neither trivial nor cheap to ignore. Storage costs money. Duplicate images pollute machine-learning training sets. They slow emergency response when drainage engineers in Itaquera or road crews in Santana are pulling visual records during a crisis.

The issue has global traction right now for a practical reason: cities everywhere expanded their sensor and documentation infrastructure rapidly during the post-pandemic infrastructure spending surge, and few built in deduplication protocols from the start. The result is bloated, redundant archives that drag on everything from AI-assisted flood modelling to routine building inspection workflows.

What São Paulo Is Actually Doing

The city's Centro de Operações São Paulo (COR), headquartered near the Anhangabaú valley in the city centre, has been testing a perceptual hashing pipeline — a technique that identifies near-identical images even when file names, timestamps or metadata differ — across its operational photo archives since the second quarter of 2025. The pilot initially covered imagery from the Sistema de Monitoramento Urbano cameras positioned along Avenida Paulista and the Marginal Tietê corridor, two of the most heavily documented stretches of city infrastructure.

Separately, IPT — the Instituto de Pesquisas Tecnológicas, the state-linked research body on Avenida Professor Almeida Prado in Butantã — has been collaborating with municipal teams on a broader data-hygiene framework intended to standardise how photographic evidence is catalogued across departments. That work is significant because São Paulo's urban data sits in silos: the flood-response agency CGEM operates different systems from the transit authority SPTrans, and neither traditionally shared deduplication logic with the urban planning secretariat.

The coordination gap is expensive. Storage contracts for municipal cloud infrastructure in São Paulo — based on state procurement filings from late 2025 — were running at roughly R$18 million annually across the major secretariats. Deduplication initiatives in comparable deployments in other cities have demonstrated storage reductions of between 20 and 35 percent, according to published case studies from municipal technology conferences in Europe. If São Paulo achieves even the lower end of that range, the annual saving would be substantial enough to fund additional sensor nodes along the city's flood-prone Córrego do Ipiranga drainage basin.

How São Paulo Compares Globally

Mexico City's Agencia Digital de Innovación Pública launched a similar audit of its Semáforo Ciudadano camera network in early 2025, but the scope was narrower — focused solely on traffic systems rather than the cross-departmental sweep São Paulo is attempting. Lagos, managing a fast-expanding urban camera network under its Smart City Initiative, is still largely in the inventory phase, without a live deduplication pipeline as of mid-2026. Bogotá's Secretaría Distrital de Hacienda has published procurement tenders for deduplication software but has not yet completed a vendor selection.

London and Amsterdam — often cited as benchmarks for municipal data governance — have had deduplication requirements written into their data procurement standards since 2021 and 2022 respectively, giving European cities a structural head start. But their systems were built with those requirements embedded; São Paulo, like most Latin American megacities, is retrofitting discipline onto infrastructure that was never designed with it.

The practical timeline matters. São Paulo's full cross-departmental deduplication rollout, if the COR pilot scales as planned, is targeted for completion in the first half of 2027. City engineers managing the drainage upgrade programme in the Várzea do Tietê area are particularly keen: duplicate imagery in flood-risk surveys has previously caused errors in elevation modelling, according to internal documentation reviewed during budget hearings at the Câmara Municipal in March 2026.

For residents and businesses that interact with municipal digital services — whether filing a building variance on Rua Augusta or contesting an infrastructure complaint in Cidade Tiradentes — the visible impact will be faster processing times and fewer errors in official visual records. The less visible payoff is a cleaner foundation for the machine-learning tools the city is deploying to predict flooding, manage traffic and allocate maintenance crews. Getting the data right, it turns out, is unglamorous work that every ambitious smart-city project eventually has to do.