What decides ANPR accuracy on Indian roads, and how to test it in a tender

How is ANPR accuracy measured?

ANPR accuracy is measured with two numbers, not one. The capture rate is the number of plates the system detects, compared with all the plates a human can see in the camera's view. The read rate is the number of detected plates the system reads correctly, compared with all the plates it detected. The UK Home Office's guidance on ANPR performance assessment (Version 2.0, July 2020) defines both terms this way.

The split matters because a single 'accuracy' figure hides where the losses are. A system can read almost every plate it detects and still miss a third of the traffic. Another can detect nearly everything and misread one character in ten. A vendor figure that does not say which measure it is, on what traffic, by day or by night, cannot be compared with another.

For a model of how to write a measurable requirement, the UK's National ANPR Standards for Policing and Law Enforcement (NASPLE, Version 3.5, July 2026) require static systems to capture 98% of plates that are visible to the human eye and meet the UK plate reflectivity standard, and to read 95% of the captured plates correctly. That is a UK standard for UK plates. It is quoted here to show the shape of a good specification, not as a figure Indian roads will reach.

How do camera placement and angle affect ANPR?

Placement decides whether the camera sees a flat, whole, readable plate. Most failures that look like software problems start here.

The UK Home Office guidance names the factors. If the field of view is too wide, plates appear too small to read. If it is too narrow, parts of the plate are cut off. Plates that are skewed or rotated to the camera, because the vehicle is turning or because the camera is mounted at a steep angle, are read less reliably, and each ANPR product has a limit on the angle it tolerates.

• Lane coverage: one camera per lane gives each plate more pixels. A camera covering several lanes spreads the same sensor across a wider road.
• Mounting height and angle: a steep downward angle distorts characters. Survey the site before fixing pole heights.
• Stand-off distance: the camera must see each vehicle long enough to capture several frames.
• Occlusion: in mixed traffic, two-wheelers and cars are often hidden behind buses and trucks. Placement should give a view between vehicles, not only down a clear lane.
• Pole stability: a pole that sways in wind or vibrates with heavy vehicles blurs the image.
• Which plate to read: specify front, rear or both, by vehicle class, because some classes carry a small or badly placed plate at one end.

How many pixels does a plate need?

A plate needs enough pixels across each character for the recognition software to tell similar characters apart, such as 8 and B, or 0 and D. What counts is the resolution on the plate, not the camera's megapixel rating.

Pixels on the plate depend on the sensor, the lens, the distance and the width of road the camera covers. The UK Home Office guidance notes that covering more lanes with one camera can reduce the resolution achieved in each lane. Each ANPR engine has its own minimum character size in pixels. Ask the vendor to state it in writing, then check it on the site's own footage.

Compression matters too. If recognition runs on a heavily compressed video stream, fine strokes on characters can be lost. Specify whether recognition runs on the camera, on a full-quality stream, or on the recorded video, and test on the same path that will be used in service.

Why do shutter speed and infrared lighting matter?

Shutter speed decides motion blur, and lighting decides whether there is a usable image at all. Both change between day and night, so ANPR has to be tested in both.

The UK Home Office guidance says the shutter speed must suit the road speed, because a vehicle that moves while the shutter is open leaves blurred characters. It also warns that a shorter shutter lets in less light and weakens low-light performance. The guidance suggests infrared (IR) illumination and suitable cameras to reduce night-time problems, and ANPR cameras commonly carry dedicated IR illuminators.

IR has its own problems. The same guidance notes that sunlight contains a lot of infrared, which can reflect off the road and cause glare on infrared cameras, and that visible-light cameras suffer glare from headlights and low sun. Illumination has to cover the whole area where plates are captured, evenly. Depth of field, the range of distance in sharp focus, must cover every frame the engine uses.

What does HSRP change for ANPR in India?

HSRP gives ANPR a more uniform target. The Motor Vehicles (New High Security Registration Plates) Order, 2001, issued by the Ministry of Road Transport and Highways (MoRTH) on 22 August 2001, sets the plate's features: letters and numerals embossed and hot-stamped, letters in English and figures in Arabic numerals, a chromium-based hologram, a laser-branded identification number in the reflective sheeting, and the colour scheme of black on yellow for transport vehicles and black on white for others.

On 4 December 2018 MoRTH notified G.S.R. 1162(E), amending Rule 50 of the Central Motor Vehicles Rules, 1989, and on 6 December 2018 it issued S.O. 6052(E), the Motor Vehicles (High Security Registration Plates) Order, 2018. Together they require vehicle manufacturers to supply HSRP, including the third registration mark sticker, for vehicles made on or after 1 April 2019, with dealers fitting them. MoRTH's advisory to states on the change also allows transport departments to authorise dealers to fit HSRP on older vehicles.

ANPR still has to handle more than one plate format. MoRTH announced on 10 August 2018 that battery-operated vehicles carry white characters on a green background, or yellow on green for transport vehicles. The Bharat series (BH-series), introduced by notification dated 26 August 2021, uses the format YY BH #### XX. An ANPR engine that was trained mainly on state-series plates should be tested on these formats before acceptance.

Which plates still defeat ANPR?

Plates that do not look like the standard still cause most misreads, and a better camera does not fix them. Some of these break the rules. The 2001 HSRP Order requires English letters and Arabic numerals, for example. But they remain on the road, and the test should count them separately so they do not hide how the system performs on legal plates.

• Stylised or decorative fonts, and uneven character spacing
• Numerals in regional scripts
• Two-line layouts, common on two-wheelers and three-wheelers
• Dirty, faded, bent, cracked or partly covered plates
• Plates hidden by tow hitches, bull bars, bicycle carriers or loads
• Temporary and handwritten registration marks
• Plates on vehicles from outside the formats the engine was trained on

How should a tender specify and test ANPR accuracy?

A tender should specify accuracy as a test method, with pass marks, rather than as a percentage in a datasheet. The method is set out before installation and run at each site.

The UK Home Office guidance recommends an evaluation sample of no fewer than 250 consecutive vehicles in the camera's view, checked against a ground-truth video read by people. Consecutive matters: a hand-picked sample flatters the system. The same guidance treats a plate as correctly read only when all its characters are right.

For enforcement, the law adds its own requirements. Rule 167A of the Central Motor Vehicles Rules, notified by MoRTH as G.S.R. 575(E) on 11 August 2021, lists ANPR among electronic enforcement devices. Sub-rule (1) requires any device used to issue a challan to have an approval certificate, signed by a designated state authority, certifying that it is accurate and operating properly, renewed every year. As notified, the rule also requires e-challans to carry clear photographic evidence of the offence and the plate, and the date, time and place. It refers to a certificate under section 65B(4) of the Indian Evidence Act, 1872. The Bharatiya Sakshya Adhiniyam, 2023 replaced that Act from 1 July 2024, so the state's challan process should be checked for how it now certifies electronic records.

• Capture rate and read rate as separate pass marks, with an exact full-plate match
• Results by lane, by vehicle class (two-wheeler, car, three-wheeler, bus and truck), and by day and night
• Ground truth from an independent video recording, read by people
• A minimum sample of consecutive vehicles at every site, not a single demonstration site
• Non-standard plates reported separately, with the reason for each miss
• Plate formats in scope: HSRP, green EV plates, BH-series and commercial plates
• Re-testing after any camera is moved or replaced, and at fixed intervals in the maintenance period
• Evidence packages that meet the state's e-challan process, and the Rule 167A approval certificate

How Ahuva approaches ANPR

Ahuva's video surveillance and ITMS line covers city traffic surveillance, enforcement, IP surveillance, AI/ML analytics and ANPR, with a centralised VMS. It entered intelligent transport, ITMS and ATCS, in 2025. It is delivering Project Sarthi for the Greater Visakhapatnam Municipal Corporation via KELTRON, whose published scope includes violation enforcement at 90 junctions and free-flow monitoring at 14 locations.