2MP cameras vs 3MP cameras

[smokingjoe 0]

I asked this question on a different post but got no replies. Recording 1080p -- 2MP camera VS 3MP camera. Do I gain anything with 3MP camera?

[ssmith10pn 0]

What kind of camera?

 

Axis P3346 is a 3Mp camera

3mp resolution is 1600 x 1200 4:3

2MP Resolution is 1920 x 1080 16:9

 

As you can see the 3mp image is narrower but taller than the 2mp image

[Securame 0]

You gain a 50% resolution increase

[ssnapier 0]

The question you have to ask is will you use it at full resolution? Do you plan to use the digital zoom features? If not, go with the 2MP since your recording storage requirements will be quite a bit lower.

[smokingjoe 0]

Thanks for the quick replies.

I install security cameras on the side in a rural area in Indiana. I'm definitely not an expert but the local people seem to be happy with what I install. I recently switch from installing anolog systems to IP cameras and using NVRs. From day one I have used Dahua systems and have had good luck with their equitment. There is a Dahua dealier located about thirty miles from where I live so this is pretty handy for me. My last installation I ordered (4) G4-IPSDE20A 2MP Dome cameras and (1) G4-IPSCX20A 2MP Varifocal camera. When I picked up the order I was told that the G4-IPSCX20A 2MP cameras were being phased out and they replaced it with a G4-IPCX30A 3MP varifocal camera for about the same price. The big difference I see according to the spec sheet is that this new 3MP camera has a Aptina CMOS image sensor and the old camera had a Sony Exmor image sensor.

I guess my question is that if I'm using a 16:9 display ratio and recording resolution of 1920X1080, should I notice any difference.

If I understand it right, which I probably don't, 1920X1080 is a little over 2 millon pixels. This is the highest recording resolution offered by the NVR. So if the new camera sends an image with 3 millon pixels to the NVR, how will the extra 1 millon pixels be able to improve the image that I'm seeing? Live or playback.

Maybe this would help when zooming in on recorded video but I don't understand how.

Edited December 31, 2014 by Guest

[MaxIcon 0]

Most cameras work this way:

 

2MP - 1920 x 1080 pixels (1080p is a standard, but there are 2MP cams that aren't 1080p)

3MP - 2048 x 1536 pixels (or whatever; 3MP isn't a standard, and actual resolution may vary; this is typical Dahua resolution)

 

3MP will show the exact same pixels as 2MP, plus additional pixels all around.

 

For the Dahua 3MP resolution listed above, you'd get 128 more pixels horizontally, giving 64 more pixels per side, and 456 more pixels vertically, for 228 more pixels each top and bottom. You don't get any better resolution in terms of pixels per foot (or whatever), just more field of view for more coverage.

 

This would give a slightly wider and definitely taller 4:3 image compared to the wide-screen 16:9 1080p image. Some cams handle this differently, like Hik's 2CD2xxx series, which have a smaller HFOV at 3MP than at 2MP, but this isn't typical. Likewise, the Axis specs listed above show that you get less FOV horizontally, and you'd have to look at the images to see if it's the same pixels as 2MP but fewer horizontally, or if it's sampled.

 

How each cam handles this depends on how the firmware is written, and you can't be completely sure until you see the actual images at each resolution.

 

Likewise, how your NVR handles this will depend on the software. It may refuse to record 3MP, it may compress it down to 2MP, distorting the image, or it may trim it to 2MP.

[buellwinkle 0]

I prefer 3MP only because I prefer the square image that's taller, so personal preference. The extra resolution mostly goes to making the image taller, so it's not going to help you ID someone any better. With most brands, you do lose frame rate when you from 1080P to 3MP.

 

The nice thing about 1080P is if you upload video to say YouTube, it will load it as 1080P. Load 3MP and it's 1080P in resolution but you lose maybe a 1/3 of the image with gray bars. So for commercial installs when they ask me to go in and find what happened, I set those cameras to 1080P so I can export them to YouTube to show them what happened and it looks good in 1080P, would not look as good in 3MP.

[smokingjoe 0]

I thank you for this helpful information. You did a good job of explaining to me how this is suppose to work. What are your thoughts on the Aptina CMOS sensor vs the Sony Exmor sensor?

[ssmith10pn 0]

Depends on the scene.

If I'm in a low ceiling parking deck I want a 16:9 Image

If I'm looking down a long narrow hallway I want a 16:9 flipped 90 degrees looking down the hallway in portrait mode.

All others 4:3 works great.

[aurmol 0]

Most cameras work this way:

 

2MP - 1920 x 1080 pixels (1080p is a standard, but there are 2MP cams that aren't 1080p)

3MP - 2048 x 1536 pixels (or whatever; 3MP isn't a standard, and actual resolution may vary; this is typical Dahua resolution)

 

3MP will show the exact same pixels as 2MP, plus additional pixels all around.

 

For the Dahua 3MP resolution listed above, you'd get 128 more pixels horizontally, giving 64 more pixels per side, and 456 more pixels vertically, for 228 more pixels each top and bottom. You don't get any better resolution in terms of pixels per foot (or whatever), just more field of view for more coverage.

 

This would give a slightly wider and definitely taller 4:3 image compared to the wide-screen 16:9 1080p image. Some cams handle this differently, like Hik's 2CD2xxx series, which have a smaller HFOV at 3MP than at 2MP, but this isn't typical. Likewise, the Axis specs listed above show that you get less FOV horizontally, and you'd have to look at the images to see if it's the same pixels as 2MP but fewer horizontally, or if it's sampled.

 

How each cam handles this depends on how the firmware is written, and you can't be completely sure until you see the actual images at each resolution.

 

Likewise, how your NVR handles this will depend on the software. It may refuse to record 3MP, it may compress it down to 2MP, distorting the image, or it may trim it to 2MP.

 

MaxIcon. Any ideas why there are more noises in the 1080p setting of the 4300s than its 3MP resolution?

[MaxIcon 0]

A bit off topic, but here are my thoughts. I haven't compared the 4300S images much recently, and have been running it in 3MP mode with a 25mm lens for a while, so not a typical install with the target well outside the IR range.

 

The Dahua directly samples sensor elements for both images, so that shouldn't make a difference. I'd expect the 2MP image to be very slightly better, since it's more from the center of the lens, which is general brighter and sharper than the edges, but the real area difference is pretty small.

 

It's also getting more of the image from the brighter area of the IR spotlight, which should help as well, but can change the exposure.

 

My best guess is that it's changing exposure times going from 3MP to 2MP, but you'd have to test it to know for sure. The Dahua will let you set a fixed exposure time, so that's what I'd do for testing - set it for 1/30 or 1/25 sec fixed (or whatever) and compare the images to see if the noise is still different. If you still see an obvious difference, post images in a new thread.

 

I don't think there's a way to read the actual exposure settings from the camera when it's set for variable exposures, but I could be wrong.

[aurmol 0]

A bit off topic, but here are my thoughts. I haven't compared the 4300S images much recently, and have been running it in 3MP mode with a 25mm lens for a while, so not a typical install with the target well outside the IR range.

 

The Dahua directly samples sensor elements for both images, so that shouldn't make a difference. I'd expect the 2MP image to be very slightly better, since it's more from the center of the lens, which is general brighter and sharper than the edges, but the real area difference is pretty small.

 

It's also getting more of the image from the brighter area of the IR spotlight, which should help as well, but can change the exposure.

 

My best guess is that it's changing exposure times going from 3MP to 2MP, but you'd have to test it to know for sure. The Dahua will let you set a fixed exposure time, so that's what I'd do for testing - set it for 1/30 or 1/25 sec fixed (or whatever) and compare the images to see if the noise is still different. If you still see an obvious difference, post images in a new thread.

 

I don't think there's a way to read the actual exposure settings from the camera when it's set for variable exposures, but I could be wrong.

 

After numerous hours of testing. This is my finding.

 

At daytime at 1/500 shutter.. 3mp and 2mp in the 4300s is almost indistiguisible.

 

But indoor at 1/30 shutter speed, there is more noise (artifacts) in the 2mp setting of the 3mp. I think the explanation is this. When they do direct sampling of the 2mp (1080p) resolution of the 3mp full sensor. Do you think they just enabled the sensor pixels at the center (the 1080p part) or all of it? It's more likely they sample the entire 3MP then crop it in software to 2mp (1080p) but during the cropping and saving it.. there is more compression introduced resulting in more noise. I did dozens of saving and comparisions indoor. I can't share sample images now because they are on private property but anyone can try them (and to those who do.. please confirm it). Thanks.

[MaxIcon 0]

If the field of view is the same for both 1080p and 3MP, they're using compression on the 3MP image, though sampled images can still have different FOVs. If the FOV is smaller for 1080p than for 3MP, they may not be sampling.

 

One way to tell for sure is to open both images in a paint program, copy the 2MP image, and paste it over the 3MP image. If it lines up exactly once you get it centered, it's reading pixels directly. If one part lines up and the rest is off center, they're sampled.

 

Here are some examples.

 

Dahua HFW3300C with 2MP overlaid on 3MP - everything lines up because both are directly sampled (note the exposures are different due to the different FOV):

 

Hik 2032 (Swann version) with 2MP and 3MP overlaid transparently - scaled images only line up in the center, then get out of alignment:

[aurmol 0]

If the field of view is the same for both 1080p and 3MP, they're using compression on the 3MP image, though sampled images can still have different FOVs. If the FOV is smaller for 1080p than for 3MP, they may not be sampling.

 

One way to tell for sure is to open both images in a paint program, copy the 2MP image, and paste it over the 3MP image. If it lines up exactly once you get it centered, it's reading pixels directly. If one part lines up and the rest is off center, they're sampled.

 

Here are some examples.

 

Dahua HFW3300C with 2MP overlaid on 3MP - everything lines up because both are directly sampled (note the exposures are different due to the different FOV):

 

Hik 2032 (Swann version) with 2MP and 3MP overlaid transparently - scaled images only line up in the center, then get out of alignment:

 

I tried the trick in the 4300s. The 2MP and 3MP images are directly taken from the sensor. But I've been thinking of something since days ago when comparing the noise.

And you mentioned the issue when you said "note the exposures are different due to the different FOV". In the camera. Do they use separate light exposure sensor for the whole image or pixel by pixel? Because if it's pixel by pixel, the light should be the same no matter what part of the sensor you take the images (whether 2mp or 3mp). Yet in the noises comparisons I did. I noticed the reason the 2mp version of the 3mp 4300s is noisier is because they use the same exposure for both even its in auto... let's say 1/30. Because the noise present are light exposure noises, not compression noise (I studied the difference between S/N vs compression artifacts). So if I adjust the exposure of the 2mp version to make it say 1/25, the noises are better or almost equal to the 3mp. But at that time I figured exposure shouldn't vary with fov because the same pixel in the sensor pick up the same light (remembering the images on the lens sides go to the sensors at the sides.. so you just get the images of the sensor at center, the light should be the same).

 

Btw.. trying the transparent trick (which I just learnt to do after years of using image editing), I found out the 1080p version of the dahua 5mp are directly taken from sensor, that is why the field of view is just one half, while that of 720p are compressed smaller from the original image (with D1 they took the center image and make it smaller). The 3mp version of the 5mp are taken directly from sensor with about 10% increase so the field of view is still half of the 5mp. I see more details at night in the 3mp 4300s than the 5mp that was why I returned it (let's ignore the 5mp for now as we focus on the exposure vs fov mystery). Thanks.

[aurmol 0]

 

Btw.. trying the transparent trick (which I just learnt to do after years of using image editing), I found out the 1080p version of the dahua 5mp are directly taken from sensor, that is why the field of view is just one half, while that of 720p are compressed smaller from the original image (with D1 they took the center image and make it smaller). The 3mp version of the 5mp are taken directly from sensor with about 10% increase so the field of view is still half of the 5mp. I see more details at night in the 3mp 4300s than the 5mp that was why I returned it (let's ignore the 5mp for now as we focus on the exposure vs fov mystery). Thanks.

 

to continue above, the following is what occurred when the dahua 5mp and its 1080p setting (in red rectangle) is overlap transparently (just learnt this trick from Maxicon.. thanks and sorry for the off topic).

 

[MaxIcon 0]

This is another case of having to figure out what they're doing in the firmware, which isn't easy, and doesn't always make sense. You have to test it to know for sure.

 

Typically, they use the entire image to calculate exposure, and when you include more pixels in an increased FOV, those pixels are now used as well. If it's sampled, the groups of pixels have different averages than the individual pixels, which can also change the exposure setting.

 

In my Dahua image, the exposure is darker at 3MP because it includes more of the sky, which is brighter, and I have auto-exposure enabled. Is your Dahua image fixed exposure on both? It appears so, because they match quite closely.

 

No telling why Dahua chose to directly sample some and scale others - this makes no sense, and causes customer confusion and unhappiness. Most sensors/processors have sampling commands built in to generate full FOV images of lower resolutions, so they didn't even have to do any special work to implement it.

 

At least now you know what's happening!

[aurmol 0]

This is another case of having to figure out what they're doing in the firmware, which isn't easy, and doesn't always make sense. You have to test it to know for sure.

 

Typically, they use the entire image to calculate exposure, and when you include more pixels in an increased FOV, those pixels are now used as well. If it's sampled, the groups of pixels have different averages than the individual pixels, which can also change the exposure setting.

 

In my Dahua image, the exposure is darker at 3MP because it includes more of the sky, which is brighter, and I have auto-exposure enabled. Is your Dahua image fixed exposure on both? It appears so, because they match quite closely.

 

No telling why Dahua chose to directly sample some and scale others - this makes no sense, and causes customer confusion and unhappiness. Most sensors/processors have sampling commands built in to generate full FOV images of lower resolutions, so they didn't even have to do any special work to implement it.

 

At least now you know what's happening!

 

The most logical is they default the exposure to 1/30 (even if it's on auto) at all low light targets to avoid motion blur. So instead of the 2Mp setting of the 3Mp needing more exposure (they just set it to 1/30). This explains all the noises of the 2Mp.

 

By the way. The same S/N noise is present in the HD-CVI camera I tried before too. So at low light, it uses 1/30 instead of say 1/10.. to avoid motion blur. Maybe the S/N in the sensor not so efficient that is why not enough signal and requiring slower exposure.

 

That said. I'd move over all this cctv now and go back to this years later when dahua produce a 12mp with incredible low light performance (if this wil exist at all in small sensors... maybe someday when adaptive optics with quantum entangled detection methods (like how the robin birds use quantum entanglement in their retinas to see magnetic fields) are used to make those tiny sensors perform like DSLR).

[MaxIcon 0]

Well, it may not seem like it to everyone, but we're already in a golden age of surveillance cams. The capabilities of $100 cams now are far greater than $1000 cams 5 years ago, and progress was much slower then. The same is true of NVRs.

 

I'm sure they'll keep improving, with the best improvements being very expensive before they trickle down to the midrange like Dahua and Hik, but low-light performance has always cost more.