Can AI Plan CCTV Camera Placement From a Floor Plan?

Yes, AI can draft a CCTV camera layout from a floor plan: it can suggest positions for entrances, tills and corridors, and flag likely blind spots. It cannot confirm what the camera will see. The installer must verify lighting, mounting height, glare, cable routes and privacy on site before anything is drilled.

Think of the AI layout as a survey checklist with sketches. It turns a plan into a short list of places to stand and look, which makes the site visit faster and the quote more precise. It does not replace the visit, because the expensive mistakes in CCTV (low sun through a glass door, a camera that looks into a neighbour's garden, a cable run through someone else's premises) are all invisible on a drawing.

Follow me on Instagram@sagnikteaches

What an AI layout is good at, and what it cannot see

A floor plan is a flat drawing of walls, doors and rooms. From that, an assistant or a planning tool can reason about sightlines: which doors face a corridor, where a counter sits relative to the entrance, which corners are hidden from a camera on the opposite wall. This is pattern work, and it is quick.

Connect on LinkedInSagnik Bhattacharya

What it cannot see is everything that is not on the drawing. It does not know that the front window faces west and will flare out the entrance camera each evening. It cannot see that a shelving unit moved last month blocks the aisle camera. It does not know the ceiling is a metal-backed tile that will not take a fixing, or that the only place to reach the cable run is through a tenant's flat. Free planning tools such as CamLayout show field-of-view cones that respect walls, which helps, but their own pages promise a layout, not a verified installation.

Subscribe on YouTube@sagnikteaches

A small shop, worked through from plan to checklist

As an illustration, take a shop of 6 metres by 10 metres. The front door is in the middle of one short wall, with a window either side. The counter sits to the right of the door, about 2 metres in. Two aisles of shelving run the length of the shop. A stockroom door is in the back-left corner, and a fire exit is in the back-right corner.

The owner wants to deter theft, record the till and be able to identify anyone who comes in or leaves. A sensible AI-drafted layout has four cameras:

• Camera 1, entrance. Mounted above the door looking in, to capture faces as people enter. This is the identification camera.
• Camera 2, till. Above and behind the counter, covering the till and the customer's hands and face across the counter.
• Camera 3, shop floor. High on the back wall, looking towards the front, covering both aisles to give an overview.
• Camera 4, rear. Over the stockroom and fire-exit corner, covering staff and delivery movements.

Now do the pixel sums. Suppose all four are 4-megapixel cameras with about 2,560 pixels across and a 90-degree horizontal field of view. At a distance d metres, the scene is about 2d metres wide. Pixels per metre is then 2,560 divided by 2d.

• Camera 1 sees people at roughly 2 to 3 metres: at 3 metres the width is 6 metres, giving about 427 pixels per metre. That is well above 250, the identification level in the 2014 edition of the IEC 62676-4 video surveillance standard, which sets detection at 25, observation at about 63 and recognition at 125 pixels per metre.
• Camera 3 looks 10 metres down the shop: at 10 metres the width is 20 metres, giving 128 pixels per metre. That reaches recognition (about 125) at the far end but not identification. At 5 metres it is 256, which just reaches identification.

So the layout works for faces at the door and at the till, and gives an overview of the floor. It does not give face-quality footage at the back of the shop from a camera mounted at the back wall looking forward. The AI draft often skips that, and a reviewer must say so. These are idealised figures. The standard's levels assume a person viewing the footage and a face about 16 cm wide, and Axis's guide to them advises a higher density as a safety margin when lighting is poor or faces are not turned towards the camera. Backlight, a steep angle, a dirty dome or a low frame rate all reduce what you actually get.

What the AI draft might say about the shop, in rough terms, and what you would fix:

Illustrative output: "Camera 1 above the entrance facing inward; camera 2 above the till facing the customer side; camera 3 on the rear wall facing the front of the shop to cover both aisles; camera 4 in the rear corner covering the fire exit and stockroom door. Potential blind spot: the end of aisle 2 behind the tall shelving near the counter."

Fixes: it did not state mounting heights, did not mention the window glare on the entrance camera, and did not note that camera 1 will also capture the pavement through the glass door. Those go on the site checklist below.

Getting a useful layout out of an assistant

A vague request gets a vague answer. Give the assistant the plan's dimensions, door and window positions, what each area is for, and what you must be able to see. A prompt you can adapt:

I am an installer planning CCTV for a shop, 6 m wide by 10 m deep.
Front door centred on a 6 m wall with a window each side. Counter on
the right, 2 m inside. Two aisles of 1.8 m shelving running the full
depth. Stockroom door back-left, fire exit back-right. Ceiling 2.8 m.
Goals: identify anyone entering, record the till, overview of the
floor, cover the rear exits. Suggest camera positions and mounting
heights. For each, say what it must capture and what could block it.
End with a list of things I must check on site before fixing anything.

The last sentence is important: asking for the on-site check list makes the assistant list its own assumptions. Read that list first; it shows you what the layout depends on.

The on-site verification checklist

This is what turns a draft into an install plan. Walk the site with the plan in hand and check each point against each camera position.

1. Stand where the camera will be and look at the scene. Is the target area visible, and is anything new in the way, such as shelving, signs, plants or a hanging light?
2. Check the light at the time that matters. Visit when the shop is open or in the evening if that is when theft happens. Look for bright windows behind the target, which turn faces into silhouettes, and dark corners with no light at all.
3. Check the sun path. Low sun through a glass door can blind a camera for an hour a day. Note which direction the door faces and whether a different angle or a wide-dynamic-range camera is needed.
4. Measure the mounting height. Too low and the camera is easy to reach and knock; too high and faces are seen from above, with foreheads and caps instead of features. A height that suits the client's ceiling and the target distance should be written down.
5. Check the mounting surface. Plasterboard, metal tile, old render or a suspended ceiling each need a different fixing. Confirm before you order brackets.
6. Trace the cable route. Where does it run from the camera to the recorder or switch? Look for fire barriers, other tenants' areas, trunking and whether a ladder is needed. Measure the length, and confirm it is inside the maximum for the cable type you plan to use.
7. Check power. Does the camera get power over the network cable, and does the switch or recorder have enough power budget for all of them?
8. Check privacy and what else is in view. Does the camera see public pavement, a neighbour's window, a toilet door, a staff break area or a changing space? Adjust the angle, use a privacy mask if the camera supports it, or move the camera.
9. Check signage position. Decide where notices go so they are seen before people enter the camera's view. For wording, see the tutorial on privacy notices and signage.
10. Test a camera temporarily. Tape one in place, view it on a phone, and walk through the area at the right distance. This catches more problems than any plan.
11. Photograph every position. Take photos from the camera's position, looking at the target. They go in the quote and later the handover pack.

That list takes about 40 minutes for a four-camera shop. It is the stage most often skipped when a layout looks convincing on a screen.

A mistake the layout did not warn about

Here is a realistic one. An installer takes an AI layout for a corner shop and fixes the entrance camera above the door, looking in. The plan was right. On the first sunny afternoon, low sun through the glass door turns every customer into a dark shape. The owner sees footage of silhouettes and phones the installer.

The layout looked correct; the failure was lighting, which the drawing cannot show. The fix is to move the camera to the side wall at a shallow angle to the door, or replace it with a camera with strong backlight compensation, and to note in the survey which way the door faces. It costs an extra visit. The checklist item on sun path would have caught it for free. A second example: an AI layout placed a camera to cover "the whole rear yard" from the back wall, which in reality overlooked a neighbour's garden. A privacy mask would have blocked that, but it should not have needed one.

Choosing lenses and heights for the shop example

The pixel sum above assumed one lens for every camera, which is rarely right. Return to the shop and look at each position again. The till camera needs a narrower view than the floor camera, because the area that matters is a counter 1.5 metres wide and the face of the person standing at it. A lens with a 60-degree field of view at 1.5 metres gives a scene about 1.7 metres wide, which on a 2,560-pixel sensor is roughly 1,500 pixels per metre, far more than needed. Going wider costs nothing there, so the till camera can take the same wide lens as the others and also catch the cash drawer and the shelf behind.

The entrance is the opposite case. People walk through a door about one metre wide, but they are in frame for a second or two, so the frame rate and shutter speed matter more than the resolution. Ask the assistant for questions to put to the manufacturer rather than for figures: what shutter setting stops blur on a person walking at normal pace, and what frame rate does the recorder store by default? Those answers live in the camera's documentation, and your own test footage will confirm them.

Heights follow a rough pattern worth writing into the checklist. For a face at the door, a mount at about 2.4 metres, angled slightly down, usually gives usable faces; at 3 metres or more, close to the door, the angle steepens and you record the tops of heads and the peaks of caps. For the till, a mount behind and above the counter at about the same height looks over the customer's shoulder rather than into their face, so an angled view from the side of the counter is often better. State the height beside each camera on the plan. If the ceiling is 2.8 metres and a camera must hang below a bulkhead, say so now.

Finally, ask the client what the footage is for. A shop wanting to deter theft needs different views from one that mostly needs to settle disputes about change at the till. An assistant can turn that conversation into a one-page brief, but only if you tell it the purpose; without one it designs for generic coverage.

A before-and-after for the client summary shows the tone to aim for. Before: "The system will catch anyone who steals." After: "Four cameras cover the entrance, the till, the shop floor and the rear exits. The entrance and till cameras are placed to capture faces in normal lighting; the shop floor camera gives an overview of activity."

Using the draft to quote and to explain to the client

Once the layout is checked, it becomes the backbone of the quote. Each camera has a position, a purpose, a height, a cable route and an estimated run length. That is the material for a line-by-line quote. For a worked example, see the CCTV installation quote template, and for handing the job over, the CCTV installation checklist and handover pack.

Clients respond to a plan with cones on it. Print the floor plan with each camera numbered and a one-line purpose beside it, for example "Camera 2: till, to record transactions and customer faces". Ask the assistant to turn the survey notes into a plain-English summary for the client, in 120 words, with no jargon. Check that it does not promise any outcome such as "will prevent theft" or "will identify every person", because no layout can promise that. A fair line is "designed to capture identifiable images at the door and the till in normal lighting".

When a free planning tool is enough, and when it is not

Free browser tools suit three cases: a homeowner deciding how many cameras to buy, an installer who wants a quick visual for a quote, and a first pass before a visit. CamLayout, for instance, lets you draw a plan, drop in camera models from a catalogue and see coverage that respects walls, and advertises itself as free. These tools come with real camera lenses, which helps with accuracy.

They are less suited to larger sites, to anything with complicated height variations such as mezzanines, or to jobs where the camera's performance is contractual. For those, use the manufacturer's design tool for the specific cameras you will install and a site survey. An AI assistant fits between the two: it is good at reasoning about a described space, flexible about odd layouts, and quick at producing a checklist, but it does not know lens data unless you give it. If you tell it the camera model's field of view, it can do the pixel sum for you; check the sum yourself.

For choosing analytics to sell on top of a layout, AI camera software for CCTV installers compares the options, and convenience store AI cameras: cost and privacy checks covers a retail case.

Three more layouts to test the method on

If you want to practise before you rely on the method, try it on three small sites where the answer is known.

A narrow terraced office. One corridor, three rooms off it, a front and a back door. The AI will probably suggest a camera at each end of the corridor. Check whether the corridor is long enough that the far end falls below the recognition level, and whether a camera facing the front door lets the street into frame.

A car-hire kiosk. A small desk, a key cabinet and a yard. The key cabinet is the item that matters. The layout must cover who opens it and when, with faces captured at the desk. Test whether the yard camera sees number plates; plates need a very different setup, with a narrow field of view and often a dedicated camera, so a general layout rarely achieves it.

A salon with a staff room. A reception desk, four chairs and a staff area. The privacy question is the interesting one: the staff area and any treatment space should stay out of view, and the AI draft may place a camera that sees into them. This is a good test of whether you check privacy rather than accept the layout.

In each, write down what the draft got right, what it missed and what the site visit found. After three sites you will know which of its habits to expect, such as forgetting heights, underestimating glare and treating every wall as clear for sightlines.

Keeping the record: what to save from each survey

A layout without a record is a guess that nobody can defend later. After each survey, save four things in the job folder: the floor plan with camera numbers, the site photographs taken from each camera position, the completed checklist with the date, and any change you made to the AI draft with the reason. If a client later asks why a camera is where it is, or a dispute arises about what the system was designed to see, the folder answers it.

Keep personal data out of the AI chat. Remove the client's name and address from any plan you upload, describe the site as "a retail unit of 6 by 10 metres" rather than naming it, and do not paste photographs that show people. If your assistant plan offers settings that control whether chats are used for training, read them before the first upload and choose the setting that suits client work.

If you would rather work through your own survey process with someone, that is what my AI implementation consultation covers; the contact page has the details, and the AI guide is a good place to see what else the same approach can do in a trades business.

Questions installers ask about AI camera layouts

Which free tools draw camera coverage on a floor plan?

Several browser tools let you draw or upload a plan, drop in camera models and see field-of-view cones: CamLayout is free, and CCTVplanner has a limited free plan and draws detection-to-identification zones from the lens data. They are good for a first layout and for showing a client. None replaces a site visit, because they cannot see glare, reflections, cable routes or what is on the other side of a wall.

Can I upload a floor plan to ChatGPT or Claude and get camera positions?

You can ask for suggested positions from a described or uploaded plan, and you will get sensible reasoning about entrances, tills and corridors. Treat it as a checklist generator. Assistants misjudge scale, miss obstructions and cannot verify pixel density, so always check the output against measurements.

How do I know if a camera will identify a face at a given distance?

Work out pixels per metre: the camera's horizontal pixels divided by the width of the scene at that distance. The 2014 edition of the IEC 62676-4 standard sets identification at 250 pixels per metre, recognition at 125, observation at about 63 and detection at 25. A 2025 edition exists, so check which one a contract cites, and allow a margin for poor light and angled faces.

Does a client's floor plan have to be accurate?

Not perfectly, but wall positions and door swings must be right, and dimensions should be within about 10 per cent. If the plan is an estate agent's sketch, treat it as a rough guide, measure on site, and redraw the key walls before ordering equipment.

Are there privacy rules I must check for placement?

Yes. Cameras that capture public space, neighbouring property, changing areas, toilets or staff break areas raise privacy issues, and signage or notice rules may apply. Rules depend on where your client operates, so confirm them and keep the camera views inside the area the client controls.

Further reads

• CCTV Maintenance Contract: Remote Health Checks and Renewals With AI — Keep the system working after installation with remote checks.
• AI Footfall Cameras for a Retail Store: What They Count and Privacy — What counting cameras collect and the privacy checks that follow.
• AI Shoplifting Detection for a Small Shop: What It Sees and the Law — What AI theft detection can and cannot do in a small shop.
• AI CCTV Cameras for a Factory: Detection, Cost and Staff Consent — Camera analytics on a larger site, with cost and staff consent.
• AI for Retail Store Design: Layout and Display Ideas — Plan a shop re-layout with AI, safely: measure, get three options, tape-test the best, check exits and access, then measure whether sales moved.
• Best AI for Shop Design: Floor Plan and Display Brief for Your Fitter — A shop owner's guide to AI for a refresh: measure, draft a layout, mock up windows and signs, then hand the fitter a brief they can price and check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: CamLayout and CCTVplanner product pages; Axis Communications white paper on pixel density based on IEC 62676-4:2014, April 2026 version (checked November 2026).