Road safety audits in India: a practical guide
Everything a network owner needs to know before commissioning one — what the exercise is, who does it, what it produces, and which parts of it a camera can now do for you.
What a road safety audit is
A road safety audit is a formal, independent examination of a road carried out by people who did not design it, to identify the features that make a crash more likely and recommend what to change. The independence is not a formality — the whole value lies in a fresh set of eyes noticing what the designer stopped seeing.
The subject can be a scheme on paper, a road part-way through construction, or a corridor that has been carrying traffic for twenty years. What stays constant is the question: given how people actually drive here, what about this road invites the wrong behaviour?
Audit versus condition survey
A road safety audit is an engineering judgement about crash risk, producing recommendations. A road condition survey is a measurement of what is physically broken, producing evidence. The audit asks why this road is dangerous; the survey asks what on it has failed. They are commissioned together and confused constantly.
| Safety audit | Condition survey | |
|---|---|---|
| Question it answers | Why might this road cause a crash? | What on this road is broken, and how badly? |
| Output | Recommendations to change the road | A classified, located defect list |
| Looks at | Sightlines, junction geometry, conflict points, speed environment | Potholes, cracking, markings, barriers, signage, kerbs |
| Performed by | A qualified independent auditor | A survey team, or an automated capture run |
| Can it be automated? | No — it is causal reasoning | Largely yes — it is detection and measurement |
| Typical frequency | At defined project stages, or after a crash cluster | Annually by tradition; monthly once collection is cheap |
One feeds the other. An audit conducted without current condition data is an opinion; condition data with nobody to interpret it is a spreadsheet. Most of what has changed in the last few years concerns the second of the two, which is why the rest of this guide spends most of its time there.
Who commissions one
Road safety audits in India are commissioned by four groups: national and state highway authorities under statutory obligation, municipal corporations managing city networks, construction firms and concessionaires proving handover quality, and fleet operators acting on duty of care. Each wants a different cut of the same underlying condition data.
- Highway authorities — statutory obligation on national and state corridors, and the need to defend maintenance budget allocations with evidence.
- Municipal corporations — dense networks, high complaint volume, and political pressure that tends to arrive faster than survey data.
- Construction firms and concessionaires — demonstrating handover quality, and defending against defect claims during the liability period.
- Fleet operators — duty of care to drivers and a direct commercial interest in vehicle damage and route reliability.
Each wants a different cut of the same underlying data, which is set out job by job on the use-cases page.
The stages
Formal audit practice recognises five stages tied to a road’s life: feasibility, preliminary design, detailed design, pre-opening, and the in-service audit of an existing road. The earlier the stage, the cheaper the fix — a conflict point removed on a drawing costs nothing, while the same correction after construction costs a rebuild.
| Stage | When it happens | What it examines | Cost of a fix |
|---|---|---|---|
| 1. Feasibility | Route and concept selection | Alignment choices, junction strategy, road type | Negligible |
| 2. Preliminary design | Layout agreed, details open | Geometry, cross-section, junction form | Low |
| 3. Detailed design | Before construction starts | Signage, markings, barriers, drainage, sightlines | Moderate |
| 4. Pre-opening | Built, not yet in traffic | What was actually constructed, driven and walked | High |
| 5. In-service | Road carrying traffic | Real behaviour, crash history, condition as it degrades | Highest |
In practice, most Indian network owners spend the bulk of their attention on the last of those, because most of the network already exists. That is also the only stage where condition data is a live input rather than a projection, and therefore the stage where automated surveying changes the economics.
The standards it answers to
Three bodies set the frame in India. The Indian Roads Congress issues technical guidance, the Ministry of Road Transport and Highways sets policy and specification, and the National Highways Authority of India governs national corridors. A survey deliverable is only useful to the extent its defect classes already map to their categories.
The practical consequence for anyone buying survey data is narrow but important: a defect list whose classes and severity grades already correspond to the categories those bodies use can go straight into a submission. One that uses a vendor’s private taxonomy has to be translated first, by someone, at cost — and translations are where defensibility leaks away. We set out how our own outputs map to those categories under validation and standards.
Where automation fits
Automation replaces the data collection that precedes an audit, not the audit itself. A camera behind a windscreen can classify and locate a whole corridor’s defects in one pass at traffic speed, turning weeks of field work into hours of review. The engineering judgement about crash risk stays with a qualified auditor.
A camera behind a windscreen, on a vehicle already making the trip, can produce a classified and located defect list for a whole corridor in a single pass at normal traffic speed. No lane closure, no specialist rig, no booking window. The output is a draft that still needs review — but review of a completed dataset is hours, where collection was weeks.
What follows from that is mostly about frequency rather than accuracy. When a survey costs a drive, you stop rationing it, and a network you can look at monthly is a fundamentally different thing to manage than one you look at annually. The four-step process shows what the operator actually does, and the public sample audit shows what comes out the other end.
Chapters in depth
Four chapters take one part of this guide further than a summary allows: how detection actually works frame by frame, how to read a vendor’s accuracy figure, where manual inspection still beats a camera, and what a finished 12.4 km audit looks like once you sort the rows.
- ExplainerHow AI detects road defects from ordinary dashcam videoHow a windscreen camera becomes a defect list: how frames turn into detections, how each one gets a location, and why an ordinary dashcam is enough.
- Case studyInside a 12.4 km audit: reading 128 defects on NH-44A walkthrough of our public sample audit: what 128 detections across 12.4 km look like once sorted, and the two-kilometre cluster that changes the plan.
- ComparisonManual road surveys vs AI video audits: an honest comparisonWhere a walked inspection still wins, where camera-based auditing wins outright, and how to decide which parts of your survey programme should change.
- DataWhat “85% pothole detection” actually means for your auditHow to read a detection accuracy figure: why 85% and 97% measure different things, why the missing 15% matters less than you think, and what it hides.
Questions we are asked
What is a road safety audit?
A formal, independent assessment of a road — either a design not yet built or one already carrying traffic — to identify features that raise crash risk, and to recommend changes. It is an engineering judgement about safety, distinct from a condition survey, which measures what is physically broken.
Who commissions road safety audits in India?
National and state highway authorities, municipal corporations responsible for city roads, concessionaires operating tolled corridors, and construction contractors demonstrating handover quality. Fleet operators increasingly commission condition data on the routes they run, though for duty-of-care rather than statutory reasons.
How often should a road be surveyed?
Traditionally once a year or less, because a manual survey is expensive enough to ration. That interval is a budget artefact rather than an engineering one — defects progress on a scale of weeks in monsoon conditions. Where automated collection has made surveying cheap, monthly is achievable and gives you rate of change rather than a single snapshot.
Can AI replace a road safety auditor?
No. Automation replaces the data collection that precedes an audit — the weeks of walking, photographing and tabulating. The audit itself is causal reasoning about why a layout produces crashes, and no detection model performs it. The practical effect is that qualified attention gets spent on judgement instead of clipboard work.
What standards do Indian road audits answer to?
Guidance from the Indian Roads Congress (IRC), the Ministry of Road Transport and Highways (MoRTH), and for national highways, the National Highways Authority of India (NHAI). An audit deliverable is far more useful when its defect classes and severity grades already map to the categories those bodies expect.
What does a road condition survey actually produce?
At minimum, a located and classified defect list: what was found, what type it is, how severe, and exactly where. Anything without coordinates is a document rather than a work order. A complete deliverable also carries the visual evidence for each finding, a map of the corridor, and an export the receiving system can ingest.