How Many Megapixels does a CCTV Design need for Identification

With megapixel surveillance cameras being available in a broad selection of resolutions, how do you know how much is enough? Many salespeople are making huge recommendations as to what to expect from a megapixel camera as an upgrade from a analogue 4CIF camera. Things like 1 camera will replace 5 cameras or even more just are not realistic for most scenes.

Take the following image, which has been provided by The CCTV Blog as an example of 1 camera replacing 16 analogue. While this does give an awesome view, at 8MP the Arecont AV8185 camera provides on this scene an overview to track peoples movement and requires addition cameras to complement it for adequate coverage.

Arecont AV8185 8MP 180 degree panoramic surveillance camera

200% zoom on 8MP image from Arecont AV8185 200% zoom from the
8MP 6292 x 1200 pixel
Arecont AV8185
image

This is the native resolution of the image, showing the artefacts from the compression.
The face is over 100 pixels high so it meets the pixel density for ID requirement but is still not of sufficient quality to truly identify this person. This may be due to the configuration and not a reflection of the cameras capability.

 

Comparing the commonly available 4:3 and 16:9 formats, we have the 4CIF as a base reference, followed by 720p HD, 1080p HD, 1.3MP, 2MP,5MP,11MP and 16MP. For each model, to make measurements consistent we have defined all cameras as being a 1/3” format using a 4mm lens mounted at 2 metres height. We are highlighting the areas that meet the Australian Standards AS4806 definition of identify and Recognise as illustrated below.

image AS4806 CCTV Identify Region

Regions by colour

ID & Recognise Regions

4CIF Resolution

4CIF is the best we have been able to achieve up until about 2007 when megapixel cameras started to hit the commercial market.

3D drawing software from Video CAD image of a 4CIF camera giving identification quality 3D drwaring software from Video CAD image of a 4CIF camera giving recognise quality
Identify Resolution 4CIF Recognise Resolution 4CIF

3D Video CAD image of a 4CIF camera giving identification qualityAt this resolution we actually struggle to get person identification that meets the standards. A person has to be standing between 1.2 – 1.5 metres from the camera for that perfect shot and the scene is only 1.5 metres wide.

The recognise region is only marginally better being 1.5 – 2.3 metres and 2.3 wide.

Because we want to be able to see under a persons hat if they are wearing one, we have defined 30 degrees as the maximum angle from the horizon, which also restricts the distance.

It is no wonder we get so few convictions on the older CCTV systems if this was the best result we could expect.

720p HD Resolution

3D Video CAD image of a 720p camera giving identification quality

3D drawing software from Video CAD showing ID quality CCTV 720p shop counter 4mm @3mtrWorking at 720p we have a 16:9 aspect ratio which gives for most applications more effective coverage as we have less roof or skyline in the scene to impact on lighting.

We can now identify a person at nearly 3m from the camera and 3.2m wide. The recognise range is out to 6m by 7m wide.

The 16:9 aspect makes covering counters ideal due to the extra width as shown here.

1080p Resolution

3D Video CAD image of a 1080p camera giving identification quality

3D drawing software Video CAD showing an image of a 1080p camera giving identification quality at 4.5mAt 1080p we have a similar result as with the 720p due to the wide aspect ratio but with a deeper ID region of 4.5m by 5m wide. The recognise region extends to past 9m by 10m wide.

1.3MP Resolution

3D drawing software Video CAD showing an image of a 1.3MP camera giving identification quality at 3m

3D drawing software Video CAD showing an image of a 1.3MP camera giving identification quality at 3m1.3 megapixels gives a similar result to the 720p with the exception of 4:3 aspect ratio so we gain height.

At just over 3m by 3m we get ID quality and recognise at double that.

2MP Resolution

3D drawing software Video CAD showing an image of a 2MP camera giving identification quality at 3.7m

2MP 4mm @3.7mID is up to 3.7m by 3.7m and as I am sure you are getting to know, recognise is double that.

The 8MP 180 degree image
above  [easyazon-image-link asin=”B0044ZRDIO” alt=”Av8180 camera (8mp, 180 feet, 6400 x 1200 and 4 8mm lens)” src=”http://ecx.images-amazon.com/images/I/21LVeKKK4xL._SL75_.jpg” align=”right” width=”70″ height=”70″]is actually 4x 2MP cameras combined to make one wide image so the same could be achieved with 4 conventional cameras with a better coverage.

3MP Resolution

3D drawing software Video CAD showing an image of a 3MP camera giving identification quality at 5m3D drawing software Video CAD showing an image of a 3MP camera giving identification quality at 5mGetting into the bigger resolutions, we start to get some significant changes in pixel density but it should be noted that we are also increasing the storage and bandwidth significantly to get the same frame rate through.

Our ID limit is now at just under 5m

5MP Resolution

3D drawing software Video CAD showing an image of a 5MP camera giving identification quality at 6.3m3D drawing software Video CAD showing an image of a 5MP camera giving identification quality at 6.3mAt 5MP many of the cameras cannot produce a full live frame rate due to lack of processing power. You may fine you can only get 5ips, which may not catch the face on the best angle for ID.

The range is now 6.3m from the camera for ID.

11MP Resolution

3D drawing software Video CAD showing an image of a 11MP camera giving identification quality at 8.3m3D drawing software Video CAD showing an image of a 11MP camera giving identification quality at 8.3m

We are pushing 8.3m for ID now but at a huge bandwidth. We are also using 11MP 35mm CMOS chipsets that are more designed for use in digital still cameras. However the results speak for themself.

16MP Resolution

3D drawing software Video CAD showing an image of a 16MP camera giving identification quality at 10.8m

3D drawing software Video CAD showing an image of a 16MP camera giving identification quality at 10.8m

At 10.8m ID coverage, the 16MP is currently the largest megapixel camera commonly available in the CCTV market but only expect 3ips at this resolution and 65Mbps bandwidth at the best quality.

Having seen these used in public spaces such as airports on a dedicated LAN I can say they are impressive as to what can be achieved and I would still recommend using several other cameras at choke points to get the person coming and leaving.

Summary

The invention of the megapixel video surveillance has definitely improved the quality of images but without the fundamental understanding of pixel density, many sites are really no better off. They have a crisper image but are trying to cover too much in many cases with just one or two cameras of a larger megapixel size but may be better off having an additional camera or 2 at a lower resolution and getting much better camera angles and typically less storage requirements.

Looking back to the original image, I can use the one camera with a 180 degree lens and save some time on the installation or I can put in six 720p cameras spread across the 17 isles.

Looking at the examples below, how would you approach this kind of installation?  Add your comments below. :-D

180 Camera 4180 Camera 3180 Camera 2180 Camera 1

imageCamera 4

180 degree camera

Image of the person in the top left of the scene shows the back of the head at a resolution that is of no use.

4mm Camera 104mm Camera 94mm Camera 84mm Camera 74mm Camera 64mm Camera 5

imageCamera 10

Individual 720p cameras

The camera alignment does not look as nice but we get an ID quality image of the same person in the far left.

Tags: , ,

10 Comments

Leave a comment
  1. Norman September 14, 2011 at 5:01 am #

    Definitely the 6 720p camera’s, much better coverage. Less obstacles of the camera are properly placed.

  2. CCTV Installer September 14, 2011 at 9:27 pm #

    Good article.

    I agree. Distance from the camera is quite important, however much you can zoom-in.

  3. SISTESEGU September 15, 2011 at 8:53 pm #

    El Número de Cámaras es también una ventaja Gigante; menos cámaras es mejor tiempo de instalación y puesta a punto del sistema. Además, mucho mas facil de manejar a la hora de buscar videos.

    The number of cameras is also an advantage, less cameras is best time of installation and tuning of the system. In addition, much easier to handle when looking videos.

  4. Vishal Kapoor September 16, 2011 at 1:59 pm #

    Really good article. Keep ‘em coming.

  5. Robin Stenzel October 6, 2011 at 5:39 am #

    Very good article.

    The reality is the Facial Recognition done via electronic processing is here and the requirements will be defined by the products used.
    The Facial recognition systems we have can work with 4CIF cameras – the only issue being how they match and how close they match. This technology has been haunted by failure because of multiple reasons; the main two are how close they match and the quality of the installation of the CCTV equipment.
    From an electronic Facial Recognition point of view the most important thing that will be faced by the CCTV industry will be the ability of installers to properly create the scene.
    Common faults that plague the industry with images are mainly warped images – exactly like that shown in this article.
    It is all fine and well placing this 4x 2MP camera in one position and covering a large area – but you have to place your head on the side to look at the images correctly on the left and the right. So you now need to ensure your software processing this image is turning its “head” on the side to correctly identify. This is a lot of code and will only drive up the costs of this technology.
    Frankly I cannot imagine the software house’s will be interested in correcting poorly installed images relating to this technology.
    In future the installers will need to be highly trained and aware of what they are doing and what constitutes a proper image – something currently not considered properly in the CCTV industry.
    Think about this, if you have a “fish eye” going on, the size of someone’s face is changing as they cross the image – this means that again someone has to “correct” the image prior to processing.
    I have seen this done I believe with a Mobotix camera and software – but are we now crossing into video tampering and will the justice system accepted tampered images that require correcting?
    Long story short – from an electronic Facial Recognition point of view you would be better off installing the 6 cameras over the 1.

    • Tim Norton October 6, 2011 at 1:19 pm #

      Robin,
      You may be correct about the legal issues of these 360 cameras because they are not giving an unprocessed image.
      Interesting to see the first big court case

  6. CCTV Installer November 6, 2011 at 1:48 am #

    I agree Robin, 360 cameras sound amazing, but in reality are not that great and complicate matters. (Apart from the cost!)

  7. Scriven King November 29, 2011 at 4:41 pm #

    What an informative article….Good stuff…..

  8. Conco February 5, 2014 at 6:31 pm #

    It has so many great security system which might be wonderful for my best safe. CCTV is one the great design for me as well. I have been more conductive by reading this info about CCTV. Thanks for letting this site..

  9. CCTV Expert May 12, 2014 at 5:51 pm #

    Yes we all love Tim and the rest of this blog. Useful stuff!

Leave a Reply