Find out why an intersection keeps crashing, not just that it does.

Article 1 · Local reporting

Johannesburg's most dangerous roads named by JMPD

Source: TimesLive, 27 August 2019 — 41 road deaths recorded in the metro in July 2019, with JMPD naming Nasrec Road, the N1, the N17, William Nicol Drive, Elias Motsoaledi Road, Klipspruit Valley Road and Moroka Road as the worst-affected corridors.

Problem. The report names which roads are dangerous but not why — no per-intersection breakdown of sightlines, signal phasing, lane geometry, or the conditions under which crashes cluster. Engineers are left to guess at root cause without a repeatable diagnostic process.

Our AI + IoT solution. Cameras, inductive-loop counters and a weather sensor at each named corridor feed Azure ML, which correlates crash timing and location against traffic volume, visibility and road geometry to shortlist which specific infrastructure factor is implicated — turning "this road is dangerous" into "this approach has an obstructed sightline during evening peak."

How the AI models solve this going forward. As more pilots come online, Claude cross-references findings against a growing base of historical accident reports across intersections with similar geometry, so recurring fault patterns (like unprotected right-turn phases) get flagged faster at newly-monitored sites, even before enough local incident history has built up.

Article 2 · National road safety data

South Africa's road death toll and the Road Traffic Management Corporation's reporting

Source: Road Traffic Management Corporation (RTMC) — the RTMC publishes periodic national and provincial fatality statistics, generally reported by province and road category rather than by individual intersection.

Problem. National and provincial statistics are useful for policy but too coarse for a road agency deciding which of hundreds of intersections in one metro to fund an intervention at first. There is no standard, continuously updated way to rank intersections by evidence-backed risk.

Our AI + IoT solution. Indlela Safe's spatial risk model treats each monitored intersection as its own case, scored on live sensor input rather than an annual provincial average, so a road safety team can rank and prioritise the intersections actually under their responsibility.

How the AI models solve this going forward. With more monitored intersections, the same Azure ML model can be retrained on a larger, metro-specific dataset, and GPT-drafted summaries can start referencing comparable interventions that worked at similar intersections elsewhere in the network — turning isolated pilots into a shared, growing evidence base.

Article 3 · International practice

Vision Zero and systemic road-safety design

Source: Vision Zero Network — an international road-safety framework holding that road deaths are preventable through safer infrastructure design, not solely through driver behaviour change.

Problem. Many road-safety programmes still focus on driver enforcement (speeding, licensing) rather than fixing the infrastructure conditions — obstructed sightlines, poor lighting, unprotected turns — that make the same intersections crash repeatedly regardless of who is driving.

Our AI + IoT solution. Gemini's imagery analysis is pointed specifically at physical, fixable infrastructure conditions — obstructions, faded markings, lighting — and every recommendation is handed to a human road safety engineer, keeping the focus on infrastructure fixes rather than driver blame.

How the AI models solve this going forward. GPT Image 2 can generate labelled concept illustrations of a proposed infrastructure fix for engineer review, helping shift funding conversations from "who was at fault" to "what would prevent the next crash here" — always as a proposed concept, never a built or approved design.

JMPD-identified fatal-crash hotspot roads, July 2019

This list is exactly what JMPD published — the source report gives the road names and the citywide monthly death toll (41), not a per-road crash count or contributing-factor breakdown. A real deployment would layer Indlela Safe's own sensor and imagery analysis on top of exactly this kind of official report to get to that level of detail.

Why the intersection-level detail below is still a demo

Getting from "Moroka Road is a hotspot" to "this specific intersection has an obstructed sightline and an unprotected right-turn phase" requires the sensor network, imagery pipeline, and per-incident crash records described in the AI + IoT stack — none of which are connected to this prototype. The watchlist and AI recommendation further down use realistic South African road names and plausible figures to show what that output would look like, clearly marked as simulated so it's never mistaken for the real JMPD data above.

Contact by department

Salah Setlhare

Director email: salah@noiraitechnologysolutions.co.za