Fisheye and Panoramic Cameras: Dewarping, Mounting and Limits of Use

Where are fisheye cameras useful, and where not? Panoramic types, pixel density by distance, camera- and VMS-side dewarping, mounting mistakes, analytics.
Technical summary
Technology
360° fisheye camera, multi-sensor panoramic camera, 180° camera, dewarping, VMS
Topic
Assessing wide-angle cameras in terms of coverage, detail, mounting and analytics
Level
Intermediate
Prerequisites
  • The concepts of field of view and pixel density (px/m)
  • Experience adding cameras and live viewing in a VMS
In this article7 sections

A single fisheye goes into the lobby, a big circle appears on the monitor and everyone is happy; until, after an incident, the question comes: “who came through the door?” Fisheye and panoramic cameras see between 180° and 360° from a single mounting point. In lobbies, shop floors or open-plan offices, they reduce the number of cameras and blind spots. In the wrong place, they produce a wide image in which no detail can be made out. Below we compare types of wide-angle camera in terms of coverage, loss of detail, dewarping and analytics compatibility, and make clear where they shouldn’t be used.

Types of panoramic camera

In catalogs, “panoramic camera” groups different architectures under a single name. In enterprise camera systems, distinguish the architecture before making a project decision; resolution, dewarping and analytics behavior change accordingly.

Type Structure Typical coverage Distinctive feature
360° fisheye Single sensor, fisheye lens 360° hemisphere from the ceiling Circular image, high distortion at the edges
180° single-sensor Single sensor, very wide-angle lens 180° horizontal from a wall Produces a panorama directly; detail falls at the edges
Multi-sensor panoramic 2–4 sensors, stitched image 180° or 360° High total pixel count; stitching limits
Multi-directional (multi-head) 2–4 heads with individually adjustable directions Flexible directions Each head a separate stream; no stitching

A single-sensor fisheye has no stitching errors, and the whole area is recorded in one stream. Multi-sensor cameras cover the same angle with more total pixels; however, double images or shifting (parallax) can occur on nearby objects where the sensors meet. A stitch line falling right in front of a door is a typical placement error on site. Multi-directional cameras are several fixed cameras in one housing; for looking in four directions at a corridor intersection, they usually give more detail than a fisheye.

Balancing coverage and detail

The limitation is simple arithmetic. The number of pixels on the sensor is fixed; as the angle widens, the pixels per unit area fall. A fisheye adds uneven distribution to this: directly under the camera is imaged at high density, and toward the edge of the image circle density falls both circumferentially and radially (in the direction away from the camera).

More megapixels don’t change this. They only push the threshold further out.

Worked example: Let a fisheye with a 12 MP sensor, an image circle about 2992 pixels in diameter and an equidistant projection be mounted on a 3 m ceiling. The target’s pixel density in the horizontal (circumferential) direction is found approximately as “the target’s distance from the image center (px) ÷ its horizontal distance from the camera (m)”:

Horizontal distance to the camera Distance from the image center Approximate circumferential density
2 m ~560 px ~280 px/m
4 m ~880 px ~220 px/m
6 m ~1050 px ~175 px/m
8 m ~1150 px ~145 px/m
10 m ~1220 px ~120 px/m

These values are upper limits obtained assuming an ideal lens and a sharp image. Radial density falls well below circumferential density at long distances; an object on the floor looks squashed toward the edge. Real lenses use different projection curves, dewarping interpolates, and night noise eats further into detail. The practical result: a fisheye can give observation and partly recognition level within a few meters around it. It isn’t an identification tool. For pixel density targets and the logic of the lens calculation, see the camera lens and focal length calculation article.

Dewarping: camera or client

Dewarping converts the fisheye image into views that show straight lines as straight. Typical views are single or double panoramas, a quad split view and virtual PTZ (ePTZ) windows. The conversion can be done in two different places, and this choice determines what’s written to disk.

Criterion Camera-side dewarping Client/VMS-side dewarping
Recorded image Corrected view(s) Raw circular image
Retrospective flexibility Fixed in the recording; you can’t look in another direction afterwards Virtual PTZ in any direction is possible on the recording
VMS dependency Low; the stream looks like an ordinary camera High; the VMS must recognize the lens
Processing load On the camera processor On the viewing station or server
Analytics Easier on the corrected view On the raw image, if there’s on-camera analytics

For forensic review, the strongest option is usually recording the raw image and dewarping it during viewing; after an incident, you can look virtually in the direction where it happened. The prerequisite is that the VMS supports the lens profile of that camera model. ONVIF compatibility gets the image into the VMS; it doesn’t on its own provide client-side dewarping support. Check the VMS manufacturer’s compatibility list by model.

A hybrid setup is also possible: the camera sends the raw circle as the recording stream and the corrected panorama as a second stream for live viewing. In this case, plan the resolution, frame rate and bitrate of the streams separately; details are in the camera multi-streaming and stream profiles article.

Ceiling and wall mounting

The mounting method directly determines image geometry. On most fisheyes, the mounting mode is selected in software (ceiling, wall, table), and the selected mode changes the dewarping algorithm. If the physical mounting and the mode in software don’t match, the corrected image comes out bent. When the complaint at commissioning is “the panorama comes out curved,” this setting is the first place to look in the console.

  • Ceiling mount: Gives 360° coverage; you get the most efficient result when it’s placed at the center of the room. Walls and doors at the room’s edges fall into the lowest-density area of the image circle. With a high ceiling, the density advantage under the camera shrinks and the edge areas move even further away.
  • Wall mount: A wall-mounted fisheye produces a horizontal panorama of about 180°. Half the circle sees the wall; half the pixels are wasted. If wall mounting is planned, a single-sensor or multi-sensor panoramic model designed for 180° often gives a better result.
  • Corner and pole mount: If 270° coverage from a building corner is needed outdoors, consider multi-sensor or multi-directional cameras. A fisheye’s low density outdoors doesn’t reach the distances where detail is needed.
  • Obstacles: A light fitting, ventilation duct or rack hanging from the ceiling costs a fisheye not one direction but a whole sector. When choosing the mounting point, check the line of sight across the whole circle with a blind spot analysis.
  • Lighting: A 360° camera also sees ceiling luminaires and windows. Bright spots upset auto exposure; don’t put the mounting point directly next to a luminaire.

Analytics and VMS compatibility

Most video analytics models are trained on examples of people and vehicles seen from the side or front in rectilinear images. In a raw fisheye image, a person appears from above at the center and tilted and bent at the edge; a general-purpose model may not work reliably on this image. Options:

  • Fisheye-specific on-camera analytics: Some manufacturers offer people counting or heat map analytics that work on the raw circle. A top-down view is an advantage in people counting; people block each other less.
  • Analytics on a dewarped stream: If server-side analytics is to be used, it should be applied to the corrected stream. Put in writing at the project stage which stream goes to analytics.
  • Virtual PTZ and events: With client-side dewarping, a virtual PTZ window isn’t associated with the recording. Configure separately which view the VMS opens when an event is triggered.

If a fisheye is to be used for people counting, verify placement and accuracy measurement with the method in the people counting analytics article. Don’t finalize the fisheye model before the VMS is chosen. If the order is reversed, it turns out at the end of the project that the camera can only be viewed as a raw circle.

Right and wrong places to use them

Field Assessment Reason
Shop floor, lobby, open-plan office Suitable General situational awareness, wide coverage with few cameras
Corridor intersection, elevator hall Suitable Seeing several directions from one point
Top-down people counting Suitable (with dedicated analytics) Little occlusion; top-down view is an advantage
Entrance door, turnstile, checkout Not suitable Not enough pixel density for identification
Plate reading, vehicle entrance Not suitable Angle and density insufficient for LPR
Long fence line, open yard Not suitable Detail falls quickly with distance
Warehouse with a very high ceiling Usually not suitable Edge areas are too far away

The setup that works in retail: a general view with a fisheye on the sales floor, and detail with narrow-angle fixed cameras at checkouts and entrances. Placement examples of this combination are in the retail store camera placement article. One question settles the decision: “In this area, after an incident, do I need to show what someone did, or who they were?” For the first, a fisheye is a powerful tool. For the second, it isn’t. For mounting height, VMS compatibility and dewarping setup, you can review the scope on our panoramic and fisheye camera solutions page.

Frequently asked questions

How many fixed cameras does a fisheye camera replace?

There’s no fixed number. For a general view, a fisheye can do the work of several wide-angle cameras; at a point that needs identification or recognition, it doesn’t replace a fixed camera. The pixel density required for each area decides.

Should dewarping be done in the recording or during viewing?

If retrospective flexibility is wanted, record the raw circular image and dewarp it during viewing; after an incident, you can look virtually in the desired direction. This requires the VMS to support the camera model and lens profile.

Is it right to mount a fisheye camera on a wall?

It’s possible; it gives a panorama of about 180°. But it’s inefficient, because half of the image circle sees the wall. For wall mounting, panoramic cameras designed for 180° usually give more detail.

Does video analytics work on a fisheye camera?

General-purpose models may not work reliably on the raw fisheye image. Use fisheye-specific on-camera analytics or analytics running on a dewarped stream, and test accuracy on site.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Video Surveillance

IP cameras, recording and central management infrastructure. We plan blind spots, retention periods and the camera network around the risks of each site.

Knowledge Center

Other articles