Skip to content

Open proposals · Policy proposal

Safe Streets Addis

Road safety as preventive medicine. A privacy-first, AI-assisted road safety enforcement design for every vehicle on the roads of Addis Ababa and other African cities. It uses what the city already owns: traffic cameras, the new number plates, digital fine payment and the electronic enforcement allowed by Regulation 557/2024.

A citizen proposal, draft for discussion. Version 0.3, 20 November 2026. By Dr. Temesgen Endalew Legesse, Board Chairman, Africa Digital Health Academy (ADHA). ADHA, through its innovation hub Tibon Hub, supported the work. The views are the author's own.

01

The problem

About 400 people die on Addis Ababa's roads every year by police count, and police data are known to under-count. Most of them are people on foot. Road traffic injury is one of the leading preventable causes of death of young Ethiopians, so this proposal treats road safety as preventive medicine: the cheapest trauma care is the crash that never happens.

WHO finds that speed contributes to around one third of road deaths in high-income countries and up to one half in low- and middle-income countries. Nine in ten pedestrians survive being hit at 30 km/h; fewer than half survive at 45 km/h.

Earlier measures show the gap. In the 2020 to 2021 limiter programme, more than 30,000 speed limiters were fitted nationally, but only 46 percent were in proper use. Checking a speed cap takes data. A system that reports tampering and links it to a consequence makes that check possible.

Sources: WHO, Managing speed (2017) (opens in a new tab); The Reporter Ethiopia, speed limiters in Addis (9 October 2021) (opens in a new tab).

02

What it proposes

Addis already has most of the parts of a modern enforcement system. What is missing is a fair, legal and trusted way to join them. The system covers every vehicle on the city's roads: private cars, minibuses, buses, trucks, motorcycles and government vehicles alike. The city can enforce speed and red-light rules fairly by joining what it already owns. In short:

  • Every vehicle, every main road. Cameras at paired roadside sites read every number plate to measure average speed and detect red-light and lane offences. Where the new national plates carry a readable chip, the chip is a second check.
  • Public transport and fleets get one extra layer. Licensed minibus taxis, buses and company fleets can also report their speed and any tampering, starting with a free phone app for coaching.
  • AI on existing cameras. A small computer at each camera turns video into coded movement data that is small enough for the 2G network. Video is fetched only for flagged cases.
  • A person checks every case before any notice is sent. Warnings come before fines, appeals are easy, and refunds are given if the system is wrong.
  • Speed limits are set centrally for each camera location: lower near schools, hospitals and markets, and changeable for roadworks.
  • Privacy by design. Only drivers with a confirmed offence are looked up, and all other records are deleted within hours. The data is used for one purpose only, and identity is held by a separate body.
  • Everything is documented in a signed audit trail, so that any later change can be detected.
  • It can start lean: a warning-only pilot on one corridor.
How Safe Streets Addis worksA six-step flow. 1, every vehicle on the road: cars, minibuses, buses, trucks and motorcycles. 2, cameras and plate readers: existing city cameras and paired roadside sites, with the speed limit set centrally for each location. 3, a small AI unit at each camera turns video into coded movement data that works on the 2G network. 4, a central system flags likely speeding, red-light and lane offences. 5, a human controller checks every case, with a second review for disputes. 6, a notice goes to the owner: a warning first, payment by telebirr or bank, and an easy appeal. Alongside the flow: taxis and fleets add a phone app for coaching, which feeds the AI unit; records of law-abiding drivers are deleted within hours; the central system sends traffic data, with congestion and closure alerts, to the Roads Authority; and a signed audit trail records steps 4 to 6.Signed audit trail, steps 4 to 61Every vehicle on the roadCars, minibuses, buses, trucksand motorcycles2Cameras and plate readersExisting city cameras and pairedroadside sites. Speed limit setcentrally for each locationTaxis and fleetsPhone app for driver coaching3Small AI unit at each cameraTurns video into coded movementdata. Works on 2G4Central systemFlags likely speeding, red-lightand lane offencesLaw-abiding driversRecords deleted within hoursTraffic data for theRoads AuthorityCongestion and closure alerts5Human controllerChecks every case. Secondreview for disputes6Notice to the ownerWarning first. Pay by telebirror bank. Easy appeal
From road to notice: machines sort the data, a person decides every case, and most records are gone within hours.

03

Privacy and fairness

All vehicle data is treated as personal data under the Personal Data Protection Proclamation 1321/2024. The system is built so that the city learns how traffic moves, and learns who is moving only when a confirmed offence requires it.

Coded identifiers
Vehicles appear only under coded identifiers that change over time. Records of law-abiding vehicles are deleted automatically within hours.
Identity held separately
The link from code to person is held by a separate body and used only for a confirmed offence, with a logged decision.
One purpose
The data is used for road safety and traffic management only. Any other use needs a separate law.
Road-facing cameras
Roadside cameras are set to capture plates and vehicles only. In-vehicle cameras face the road, and footage stays in the unit.

Fairness

  • A trained human controller confirms every flagged case against written criteria, and a second reviewer checks disputed ones. Error rates are published.
  • The first months are warning-only.
  • Payment is by any approved channel, including telebirr. An appeal pauses payment, and a wrong fine is refunded.
  • All vehicles are treated equally, including government vehicles. Any exemption must be listed in law.
  • Operators, controllers and suppliers are paid a fixed fee, independent of the number of fines. Running costs come from a fixed budget line.

European courts accepted average-speed control only where a specific law set these limits. The proposal therefore asks that the deletion rule and the purpose limit be written into the regulation itself.

04

What this could mean for the public bodies

These are respectful suggestions from a citizen, offered for discussion.

What this could mean for the Addis Ababa City Transport Bureau and City Roads Authority

The proposal suggests the Bureau as the lead agency, to be confirmed by the City, and the Roads Authority as the main user of its planning data. This could mean:

  • Leading the system, if the City so decides, with a control room at the city's traffic control centre where trained controllers review every flagged case.
  • Choosing one pilot corridor and starting with the cameras the city already owns, warning-only throughout.
  • Setting speed limits centrally for each camera location, including lower limits near schools, hospitals and markets, and temporary limits for roadworks.
  • Using the coded traffic data for planning: predicting congestion, timing roadworks and warning drivers of planned closures days ahead.

What this could mean for the Ministry of Transport and Logistics

The proposal rests on national rules that the Ministry leads. This could mean:

  • A clear rule for camera evidence and owner liability under Regulation 557/2024, so that a notice sent from a camera rests on firm legal ground.
  • A certification and calibration route for roadside devices, agreed with the national metrology body, so that only certified, regularly calibrated equipment is used as evidence.
  • Support for a decision by the City and the Ministry of Finance on where fine revenue goes, for example a city Road Safety Fund for crossings, footways, lighting and school zones.
  • A link to the national point-based penalty record, so that points go to a driver who is named and does not contest, or who is identified by evidence.

What this could mean for the Ethiopian Artificial Intelligence Institute

The proposal suggests the Institute as the home of the technical capability. This could mean:

  • Leading the AI models with Ethiopian engineers, so that the national capability is owned in Ethiopia.
  • Testing the models on Ethiopian roads, in Ethiopian light, traffic and weather, before any notice depends on them.
  • Publishing the accuracy and error rates of the models, so that the public and the courts can check them.
  • Building on an open-source stack, so that other African cities can reuse and audit the work.

05

The lean pilot and costs

The system can start small. A lean, warning-only pilot on one corridor, using existing cameras and drivers' own phones, is proposed to run for 6 to 9 months.

Lean pilot, cash only
USD 27,000 to 125,000
One corridor, existing cameras, phone app, warning-only. The cash comes on top of staff time, existing cameras and hosting given in kind. Import duty is extra.
Full pilot
USD 2.0 to 4.8 million
24 roadside sites covering all vehicles, plus 2,000 sealed units for public transport.
Full city-wide system
USD 8.4 to 44.0 million
Capital cost, estimated. Fines would cover only about 40 to 60 percent of its annual cost.

These are indicative estimates at 160 birr per US dollar, to be replaced by quotes. The pilot will measure whether the lean path earns more in fines than it costs. If there is a surplus, the proposal is that it goes to a city Road Safety Fund.

The proposed path

  1. Agree (1 to 3 months). Name a lead agency, and review the law on camera evidence, owner liability and data protection.
  2. Lean pilot (6 to 9 months). One corridor, all vehicles on it, existing cameras, plus 200 volunteer taxi drivers with a coaching app; warning-only throughout.
  3. City pilot (12 to 18 months). Three corridors and a published evaluation.
  4. City-wide (2 to 3 years). All main corridors, by a city decision.

06

Expected impact (modelled)

Once the system covers the whole city, the central estimate is about 40 deaths and about 190 serious injuries avoided each year. The range is wide, and these are modelled estimates.

Modelled yearly impact once the system covers the whole city
Once city-wideLowCentralHigh
Fall in average speed2 percent5 percent8 percent
Deaths avoided per year84097
Serious injuries avoided per year56194372
Value of harm avoided per yearUSD 1.4 mUSD 5.8 mUSD 12.4 m

The model links a fall in average speed to fewer deaths using the established power model, which was built on rural roads and fits city streets less well. The pilot will measure speed and injury crashes; a change in deaths can be proven only at a larger scale. Police counts are known to miss deaths, so the true benefit is likely higher. Enforcement is one part of a Safe System, alongside pedestrian crossings, footways and speed-calming works.

07

The full proposal

The full proposal is a concept note, a technical annex and a cost model, free to use and adapt under CC BY 4.0 with credit to the author.

To read the full proposal, write to info@drtemesgen.com, or open the folder below and press "Request access" (a Google account is needed). The author answers each request by e-mail with the files. Your name and address are used only to answer your request and stay with the author.

Licence: Creative Commons Attribution 4.0 (CC BY 4.0) (opens in a new tab). Credit the author.

08

Pilot, adapt or implement this design

Public bodies, partners and researchers who wish to test, adapt or implement this design, starting with a lean pilot, are welcome to get in touch.

Limited availability; reply within 2 working days.