Skip to content

Camera technology

SD vs HD vs 4K: choosing the right resolution for security cameras

SD vs HD vs 4K for security cameras: what each means in megapixels. Why pixels per metre matter more, the storage maths and what AI needs.

Updated · Technically reviewed by the Visylix Engineering Team

Wide stone city square with a fountain and street lamps at blue hour

Security camera resolution is how many pixels a camera captures, and the right choice depends on what that camera must see, not on the biggest number.

Most security cameras need Full HD (1080p). Use 4K only where you must identify faces or plates far away. For the basics of the cameras themselves, see what an IP camera is. We do not sell cameras, so this advice is brand neutral.

SD vs HD vs 4K: the right question

Most resolution discussions in surveillance start with the wrong question. Buyers ask whether 4K beats 1080p, vendors compete on pixel counts, and specifications get approved that the storage budget cannot sustain six months later.

The right question is not “which resolution is best?” It is “which resolution is right for each camera?”, given its scene, its job, the storage budget and the site's bandwidth.

This guide explains SD, HD, Full HD and 4K, what they mean for surveillance, how resolution affects faces, number plates and investigations, the storage and bandwidth maths, and how to design a mixed estate that works.

What SD, HD, Full HD and 4K mean

Security camera resolution is the number of pixels in each frame. SD is a third of a megapixel. HD is about one. Full HD is two, and 4K is about eight. More pixels show more detail. They also need more storage and bandwidth, so match each camera to the job it must do.

Resolution is the number of pixels in the picture, written as width by height.

SD (standard definition): 720x480 pixels (NTSC) or 720x576 (PAL). The norm for analogue CCTV before IP cameras. Essentially obsolete for new installations today.

HD (720p): 1280x720 pixels. The first widely sold IP camera resolution, still found on tight budgets.

Full HD (1080p, sometimes called 2K): 1920x1080 pixels. The most widely used surveillance resolution and the right default for indoor cameras, mid range outdoor cameras and most general work.

4K (Ultra HD): 3840x2160 pixels, about four times Full HD. Increasingly the default for wide outdoor areas, number plate capture, perimeters and any scene where you need to zoom into detail later.

8K (7680x4320) exists but rarely makes sense for surveillance, because the storage, bandwidth and processing costs seldom beat a well placed 4K camera.

How many MP is Full HD?

About 2 megapixels: 1920 by 1080 pixels is 2.07 million pixels.

How many MP is 4K?

About 8 megapixels: 3840 by 2160 pixels is 8.3 million pixels.

What is the resolution of a 3MP camera?

2304 by 1296 is common, but makers vary, so check the specification sheet.

Is 48MP the same as 4K?

No. A 48 megapixel phone sensor has far more pixels than 4K video, which is about 8 megapixels. Phones combine pixels when they record video.

Which is best, 2K or 4K CCTV?

It depends on distance and task. 2K (about 4 megapixels, often 2560 by 1440) is a strong step up from 1080p for medium scenes. 4K is worth it for wide areas where you must identify a face or read a plate far from the camera. For a corridor or a doorway, 2K or even 1080p does the job with less storage.

What is the difference between a 2MP and a 4MP CCTV camera?

A 2MP camera records Full HD, about 2 million pixels. A 4MP camera records about twice as many, usually 2560 by 1440. Over the same scene, 4MP puts more pixels on each person, so detail holds up further away. It also needs more storage and bandwidth, roughly in proportion.

Pixels per metre: what really matters

Pixel count alone is not what matters. What matters is pixels per metre on the subject at the distance the camera is mounted. A 4K camera covering a 100 metre car park puts fewer pixels on each person than a 1080p camera covering a 10 metre doorway. On the doorway, the 1080p camera wins, despite a quarter of the pixels on paper.

The international standard IEC 62676-4 sets widely used reference densities for different tasks:

Detection (is that a person?): about 25 pixels per metre.

Observation (what is happening?): about 62 pixels per metre.

Recognition (is that someone I know?): about 125 pixels per metre.

Identification (who is it, beyond reasonable doubt?): about 250 pixels per metre.

Number plate capture needs enough pixels across the plate for every character to be legible, and dedicated ANPR cameras are designed to deliver exactly that.

These densities decide whether a resolution is right. A 1080p camera with the right lens at the right distance can capture identification level detail on a face. A 4K camera with the wrong lens at the wrong distance may only manage observation. Resolution is necessary, but not sufficient.

Pixels per metre matter more than the resolution on the box. The standard IEC 62676-4 links each task to a pixel density at the target. Spotting that a person is present needs far fewer pixels than identifying a face. A 1080p camera with the right lens often does the job that buyers expect to need 4K for.

SD vs HD vs Full HD vs 4K for surveillance

Here is what each resolution can really do.

SD (720x480 or 720x576)

Obsolete for new installations. Too few pixels for reliable faces at any reasonable distance, usually analogue and needing conversion for modern platforms, and no longer meaningfully cheaper to store than HD. SD belongs only in old systems awaiting an upgrade. Plan to replace it anywhere footage is used for security, compliance or evidence.

HD (1280x720)

The tight budget option. Enough for general awareness, motion detection and watching at reasonable distances. Weak for faces beyond a few metres and for number plates. With Full HD cameras now so affordable, HD rarely represents the best value. It suits low priority indoor cameras, such as corridors and storerooms, where the job is watching rather than identifying.

Full HD (1920x1080)

The right default for most security cameras. Enough for face recognition at typical doorway distances, number plate capture at entry points with the right lens, and clear awareness across general scenes. Its storage and bandwidth needs are well understood and supported everywhere. Start new estates at Full HD and step up to 4K only where scene size, identification needs or zoom work justify it.

4K (3840x2160)

4K is right where the scene is large, identification matters at distance, or operators need to zoom into part of a wide view. Typical uses: long perimeter fences, large car parks, number plate capture at entry lanes, and any camera where AI or operators must pull detail from part of the scene without moving the lens. 4K is heavy on storage and bandwidth, and the cost multiplies across a large estate. Specifying 4K everywhere is the most common cost mistake in surveillance buying.

The storage and bandwidth maths most buyers get wrong

Storage grows with resolution, frame rate, compression, recording mode and retention. Most teams underestimate at least two of these, and the storage budget becomes a problem six to twelve months after go live.

The arithmetic is simple. Storage per day equals bitrate in megabits per second, times 86,400 seconds, divided by eight, giving megabytes. A camera averaging 3 Mbps uses about 32 GB a day; 8 Mbps about 86 GB; 12 Mbps about 130 GB.

Worked example: 100 cameras at 30 days retention. If they are 4K averaging 8 Mbps, that is about 260 terabytes. If they are Full HD averaging 3 Mbps, about 97 terabytes. At 100 cameras the difference is significant; at 1,000 cameras it is the whole budget conversation.

Three choices change the maths substantially. Event based recording: lower bitrate most of the time, full quality when motion or an AI event occurs. Efficient compression: H.265 uses far less storage than H.264 at similar quality. Tiered retention: recent footage on fast storage at full quality, older footage moved to cheaper storage.

The costliest mistake is continuous 4K recording across an entire estate without modelling storage at full size. The better approach mixes resolutions by scene, then applies event based recording, H.265 and tiered retention.

Resolution drives bandwidth for live viewing and site to site links, not just recording.

A Full HD camera streaming live with H.265 typically uses a few megabits per second; a 4K camera, several times that. Across dozens or hundreds of cameras, live viewing can exceed what the network was designed for.

The approach that works: record full quality locally, and use each camera's lighter substream for routine live tiles, switching to the full quality main stream when an operator opens a camera full screen. Recording resolution and viewing bandwidth are then independent, so you can record in 4K without every viewer pulling 4K.

Remote sites are harder still. Fifty 4K cameras can generate more traffic than most site internet links can carry. Record locally and share selectively: only the cameras and time ranges that matter.

Recording in H.265 roughly halves storage at any resolution; see HEVC vs H.264 and video bitrate explained.

Choosing the right resolution for each camera

Resolution is a per camera decision. The best estates match each camera to its scene and its job.

Step 1: Decide each camera's job: detection, observation, recognition, identification or number plate capture. Use the IEC 62676-4 densities above.

Step 2: Measure the scene: distance to the subject and the width the camera must cover at that distance.

Step 3: Calculate pixels per metre: horizontal resolution divided by the width covered, in metres, at the working distance.

Step 4: Match resolution to need. If 1080p with the right lens reaches identification density, 4K adds cost without value. If 1080p only reaches observation where you need identification, choose 4K.

Step 5: Prove it with a pilot. Calculations work on paper; real scenes have lighting and weather that calculators miss. A pilot on the actual scene with the actual lens confirms the design.

A mixed resolution estate that works

Most enterprise estates are mixed resolution by design. A typical large commercial site:

Entrances, lobbies and anywhere faces must be identified: Full HD with the right lens, or 4K where the scene is wide.

Corridors and general internal watching: HD or Full HD, with HD only if the saving matters at scale.

Car parks, perimeters and wide outdoor areas: 4K, often multi sensor 4K cameras for very wide scenes.

Number plate capture at entry lanes: 4K or dedicated ANPR cameras.

PTZ cameras: increasingly 4K, where extra pixels help zoom work.

This delivers identification detail where it matters, controls storage and bandwidth where it does not, and avoids both under specifying critical cameras and over specifying the whole estate.

How resolution affects AI analytics

AI is sensitive to resolution, but not in a straight line. Doubling resolution does not double accuracy, and different tasks respond differently.

Face recognition improves with resolution until the face carries enough pixels, then gains level off.

Number plate recognition is the most resolution sensitive task. Accuracy falls sharply when characters are too small, which is why 4K and dedicated ANPR cameras are standard at entry lanes.

Object detection (people, vehicles, packages) is tolerant. Full HD performs strongly across most scenes, and strong AI detects small, distant subjects even in wide views.

Pose estimation and unsafe behaviour detection benefit from enough detail to see posture. Full HD is usually sufficient; 4K helps for wide scenes where workers are far away.

Crowd detection and counting are tolerant for overall crowd levels but need more detail to follow individuals through a crowd.

So let each camera's AI job guide its resolution. A door camera for face recognition benefits from Full HD or 4K with the right lens; a lobby camera only measuring crowd levels does not.

On Visylix, AI comes with Enterprise, on the cameras you choose: see AI analytics on the cameras you choose.

Common resolution mistakes

4K on every camera. The most common mistake. Storage and bandwidth multiply across the estate, most cameras do not need the pixels, and camera spending crowds out the storage, network and software that decide whether the system works.

HD or SD to save money. The other side of the same mistake. Under specifying critical cameras, such as entrances, identification points and plate cameras, saves money at purchase and loses it the first time footage cannot identify anyone.

Ignoring the lens. Resolution and lens together set pixels per metre. A 4K camera with the wrong lens can deliver less useful detail than a Full HD camera with the right one.

Ignoring frame rate. Doubling frame rate roughly doubles storage. Most surveillance does not need more than 25 frames per second, and general watching can often run lower.

Ignoring compression. H.265 uses substantially less storage than H.264 at similar quality.

Ignoring retention. Storage grows roughly in line with retention: 90 days needs about three times the storage of 30 days. Model it before defaulting to long retention.

Camera shape matters as much as resolution; see our guide to CCTV camera types.

How Visylix gets the best from every resolution

Visylix is an automated AI video surveillance platform built in India, designed for real, mixed estates. It runs SD to 4K cameras side by side, recording each at its native resolution, and watches them with 22 AI analytics built in house.

For most sites, Full HD is the right default. 4K is worth it only for wide scenes where you must identify faces or read plates far away. Recording in H.265 roughly halves storage at any resolution. Live tiles can use a lighter second stream, so a 4K camera does not slow the video wall.

Full quality recorded, light streams viewed. Visylix records H.264 and H.265 exactly as each camera sends them. Small live tiles on video walls of up to 36 tiles use each camera's lighter substream, so an estate recording in 4K does not have to stream 4K to every viewer. Live view arrives in under a second over WebRTC, down to under 200 ms.

Storage under control. Event based recording, driven by motion, AI detections or operator triggers, keeps storage lean.

AI that makes 4K pay off. On Enterprise, on the cameras you choose, Visylix applies face recognition, number plate recognition, object detection, pose estimation and more to the stream each camera natively produces, and detects small, distant people and objects that simpler systems miss, so the extra pixels in a 4K camera genuinely improve detection.

Licensing that leaves the decision to you. Visylix ships as a Docker image on your own hardware, on premise, at the edge or fully air gapped. Starter and Pro include unlimited cameras on the VMS licence, from ₹4,999 or $49 per month, so choosing resolution is purely an operational decision, never a licensing one. AI comes with Enterprise, on the cameras you choose.

Designing a new estate or hitting the storage wall a 4K rollout often meets? Talk to the Visylix team.

FAQ

What is the difference between SD, HD, and 4K?

SD (standard definition) is usually 720x480 or 720x576 pixels, found mainly in old analogue cameras and obsolete for new installations. HD is 1280x720 pixels: fine for general watching, weak for faces and number plates at distance. Full HD (1080p) is 1920x1080 pixels and the right default for most modern security cameras. 4K (Ultra HD) is 3840x2160 pixels, about four times Full HD, ideal for wide areas, number plate capture and zooming into recorded detail.

Is 4K better than 1080p for security cameras?

Not everywhere. 4K wins for wide scenes where you need to zoom into detail later, for number plate capture at entry lanes and for long perimeters. 1080p is as good, or simply sufficient, for indoor cameras, doorways and any scene where the camera is close enough that 1080p already captures identification level detail. Specifying 4K on every camera is the most common cost mistake in surveillance buying.

What resolution is best for security cameras?

It depends on the scene, the job and the mounting. Full HD is the right baseline for most cameras; 4K is right for wide areas, number plate capture and zoom heavy work. The real question is which resolution delivers the pixels per metre the job needs at the working distance. The IEC 62676-4 standard links detection, observation, recognition and identification to specific pixel densities.

How much storage does a 4K security camera use?

It depends on bitrate, which depends on scene activity, frame rate and compression. As simple arithmetic, a 4K camera recording continuously at 8 Mbps uses about 86 GB a day; at 12 Mbps about 130 GB. A Full HD camera at 3 Mbps uses about 32 GB a day. Multiply by cameras and retention days for the total. Event based recording, H.265 and tiered retention all cut this substantially.

What is 1080p vs 4K?

1080p (Full HD) is 1920x1080 pixels, the most widely used surveillance resolution. 4K (Ultra HD) is 3840x2160 pixels, about four times as many. 4K shows more detail at distance and lets you zoom into recordings far more, at the cost of much more storage and bandwidth. 1080p suits most general cameras; 4K suits wide areas, number plate capture and zoom heavy work.

Is HD resolution good enough for security cameras?

HD (720p) is acceptable for general watching on low priority cameras, such as internal corridors and storerooms, where faces and number plates are not needed. It struggles with faces beyond a few metres and with number plates at typical entry distances. For identification or AI that needs detail, choose Full HD or 4K.

Why are pixels per metre more important than resolution?

Resolution is the total pixels in the picture. Pixels per metre is how densely those pixels cover the subject at the working distance, which decides what you can actually see. A 4K camera covering a 100 metre car park puts fewer pixels on a face at the far end than a 1080p camera covering a 10 metre doorway. Resolution is necessary, but pixels per metre at the working distance is what counts.

How does resolution affect AI analytics?

Differently for different tasks. Number plate recognition is the most resolution sensitive and benefits most from 4K or dedicated ANPR cameras. Face recognition needs enough pixels on the face, with diminishing returns beyond that. Object detection, pose estimation and crowd detection are more tolerant, with Full HD usually enough at typical distances. Let each camera's AI job guide its resolution, not the other way round.

Should I replace my old SD cameras with 4K?

Not automatically. Replace each SD camera with the resolution its job needs. Cameras for identification or AI benefit from Full HD or 4K, depending on the scene. Cameras for general watching can move to HD or Full HD without the extra storage cost of 4K. Design the new estate camera by camera rather than upgrading everything to the highest resolution available.

What is the best codec for 4K surveillance?

H.265 (HEVC) is the standard choice for 4K surveillance, using substantially less storage than H.264 at similar quality. H.264 remains the most universally compatible but is no longer the best choice for new 4K installations where storage cost matters. AV1 is emerging in streaming but is not yet common in surveillance cameras.

Sources

  1. ITU-T H.264 Advanced Video Coding
  2. ITU-T H.265 High Efficiency Video Coding
  3. ONVIF Standard

This guide supports Live view in under a second on the Visylix platform.