I thought it would be fun to run the same tests on the new Pi cameras that I did on the original one back in May 2013. Unfortunately my first Pi camera 2.1 was faulty so I had to wait an extra day before I could do it. (It’s the first time out of about 50 Raspberry Pi products that I have ever had a faulty one.) But now here are the results. I’ve compared both 2.1 cameras with both 1.3 Raspberry Pi cameras.
Methodology
I shot the scene out of an open window, trying to get all three cars and their registration plates into view. All four cameras were used, one after the other as quickly as I could swap them over. The latest camera software was used in the default maximum resolution for each camera. I had to do some vertical and horizontal flipping, depending on which camera I was using.
The idea was to try and get the same conditions for each shot. It was partially cloudy though, so not ideal weather. But what we have serves the purpose for a quick comparison. The 2.1 shots are larger because they are 8 Megapixels vs the 1.3’s 5 Megapixels. You can click the image to get a 1500px version…
Wider Field Of View – Bit Less Exposure
The Pi camera 2.1 seems to give a nice colour balance compared to 1.3, but this shot appears a little dark and underexposed. It may be because of the wider field of view and more sky in the shot. Auto-exposure is a bit of a black art. It will be interesting to see what happens in other scenes and lighting conditions.
This seems to be consistent as we got the same difference between the NOIR versions too. This is hardly surprising since they’re the same cameras without IR filter. But it gives us a nice cross-check.
I also shot the scene with my Nikon D90 on an auto mode (aperture priority at f/8) to compare. I normally shoot exclusively in manual mode with this camera but thought it would make an interesting comparison. Obviously the sharpness of a £400 DSLR is in a completely different league, but it gave an exposure result in between the two Pi camera models…
What About Resolving Power At Distance?
When I tested the Pi camera 1.3 it was capable of resolving a car registration plate across the street. Fortunately my neighbour hasn’t changed vehicles since then, so I thought I’d do the same again. That’s the reason for the choice of shot – not because the view out of the window is particularly fabulous.
So here I’ve cropped in on the parts containing the three cars. You can click the image for the 1500px version…
If you want the full resolution (6 MB) version, click here.
Now Let’s Zoom Right In to 200%
To make it easier to compare resolving power, I’ve gathered together and enlarged the registration plates to 200%. If you click the following image you’ll get the full-size version…
In the above less magnified shot, the Pi camera 1.3 looks a little clearer. If you click the above photo to go to the fully magnified shot, you’ll see that the newer Pi camera has slightly more resolution. The 1.3 camera looks a little “blocky” for the first registration plate. It’s similar for the middle one, but the one on the right is not very sharp on the newer camera. For that one, the older camera is noticeably clearer. Some lenses perform better around the periphery than others. It looks as if the 1.3 has a slight edge here on ‘peripheral vision’.
Both the NOIR cameras were a little less sharp on the registration plates compared to their IR-filtered couterparts. I imagine this is the ‘nature of the beast’ and is due to the IR hitting the sensor.
Lots More Testing To Do
This is just one use-case and one set of lighting conditions. There’s a lot more playing to be done. I look forward to seeing the output from other people’s cameras.
Nice to see these! The new camera delivers darker and (more importantly) softer images.
While you can afterwards change the dark on stills and images, the soft nature is a disappointment. It may be due to firmware not being optimal yet.
Sharpness/focus at infinity seems a real issue with the V2 camera, more reports are coming in, see the forum https://www.raspberrypi.org/forums/viewtopic.php?f=43&t=145815
Up to 5m fine, beyond that less focus. Adjusting the lens helps a bit for some, but it is not that easy to unlock the screwed on lens as in V1.3.
It would seem that the 1.3 version is not as dark as the 2.1 which makes it more ledible.
I am undecided on the 1.3 or 2.1 issue.
I would like to have seen a night-time – using with and without (noise detection) an IR floodlight – test for the IR Cameras.
There’s plenty of scope for more testing. I don’t have an IR floodlight though. I believe AverageMan is going to do some testing of the NOIR camera with his security camera setup.
Yep – going to have a play with a new NoIR and 9x IR LEDs of differing values. Not sure if 9 LEDs is really enough, but that’s what I’m starting with.
Gotta start somewhere :)
Interesting, thanks Alex!
With the increased field-of-view on the new cameras, I wondered if the extra 3 mega-pixels accounted for “more” than just the extra FOV, and they do :-) Here’s the 1.3 image overlaid at 70% opacity on top of the 2.1 image, with the two images aligned on the black car: https://dl.dropboxusercontent.com/u/340607/RaspiTVImagesOverlaid.png (all copyright remains with Alex Eames). The 1.3 image is “smaller” than the 2.1 image, which means in theory the 2.1 images contain more detail in the areas covered by both images, IYSWIM.
I think the autoexposure on the 2.1’s is correct and the 1.3’s are over exposing.
Look at the white van, for example. In the 2.1, you can see detail on the side. In the 1.3, it’s blown out. Same for the sky — there’s contrast and detail in the 2.1’s clouds. The 1.3 has tons of straight white. More overexposure.
Also consider that on the non-Noir shots, you have a lot more clouds present in the 2.1 shot. This would cause the autoexposure to have a darker image as it tries to calculate middle grey for the whole scene. Even with less extreme dynamic range, the 1.3 has a harder time properly exposing.
These look good to me.
To be honest, Mike, I think the Nikon D90 is correctly exposing, which is why I included it. I think the 1.3 is over-exposing and the 2.1 is under-exposing. The problem is, if you underexpose by too much, you lose a lot of the information. (Expose to the right and all that https://photographylife.com/exposing-to-the-right-explained )
I agree with you about the proportion of sky in the photo, that’s why I mentioned it…
Hello,
I am a bit dissapointed by the picture’s quality of the PiCam 2.1, picture sharpness is far away behind V1.3, averything look blurry compared to 1.3, look at the back of the maintenance car that is the most visible difference but if you look at full picture there is no place were you get more details on 2.1 compared to 1.3 that is sharper and finally give more details with it’s 5MP than 8MP :-(
As I keep telling people who aren’t “photography nutters”, more megapixels don’t automatically make better cameras. The lenses are far more important. To be fair to the new camera though, it does have slightly more resolving power, but the “across the frame sharpness” on my example doesn’t seem to be quite as good. It looks like an optical issue to me.
I’m sure we’ll see some software improvements as time goes by. We certainly did with the previous camera. I hope there is some scope for improving the optics as well. I don’t know if it’s lens quality or alignment or a manufacturing issue. Or if it’s just the nature of the beast. Time will tell when some other people post their results.
I am aware about the fact that lens are very important part that can ruin the best CCD and right now the result for this 2.1 is bad if I look at your samples.
And if it is a lens issue there is no software that will fix that because whenever they apply some sharpening filter it will never give back the details loss, there is a hudge difference between “sharpening ” and “details”, sharpening a soft picture will never give details back.
I own two PiCam V1.3 bought at same time (so should be same production serie) and both do not give same result at all one is “sharp” as expected the other is bad and produce softer images (strangely in the middle of the picture).
Also the wider is the lens the more you will have to apply software correction to re-align the distortion but often there is also corner blur effect with many lens because of the size of the lens againt the size of the CCD the smealler is the lens the more corner effect you’ll get.
But the fastis when you by a camera its a game sometimes you have a well balanced quality, sometimes not.
I think I’ve had about 5 Picam 1.3 and they’ve all been sharp enough. Admittedly I haven’t run the same test on all of them though. Whether or not the camera for an educational computer needs to be as sharp as that is another thing. But for use as a security camera, the ability to read a car number-plate across the street is a definite plus. We can’t ask the criminals not to park on the extreme right-hand side of the shot :)
On the side-by-side picture of the 4 PiCams modules I noticed that the screw’s lens turn position is not the same on all modules and as it’s what set the focus, may be it’s the cause of the problem ?
Hey ! on the french Blog “LaFramboise314” (raspberry314 in english) the guy received its 8MP V3 and also get dissapointed about the picture’s sharpness far behind the old 5MP but he played a bit the the lens screw position and … tadaaaa, picture went from blurry to sharp !
So my thought was good, I think there is a problem at RaspiCam factory, they do not verify the position of the screw and as the screw seems to be “randomly positonned” I suppose it is a human action.
I’m a bit disappointed with my Picam 2.1. My initial impression is that it doesn’t seem anywhere near as sharp as the 1.3. I’ll do some more experimenting over the week-end.
Have a look at the low light performance too, as I’ve heard the new camera is far better indoors with the lower light levels.
I’d rather slightly fuzzy but actual contrast in my photos than dark blobs with sharp edges.
You’d expect that from a newer generation sensor with better high ISO performance. I’m hoping to see some other people publish photos soon to see how they compare with my experiments.
I posted some links to further tests on the Raspberry Pi forum here: https://www.raspberrypi.org/forums/viewtopic.php?p=965459#p965459
I’m removing the v2.1 from my TrailCam and replacing it with the V1.3 that I took out. The v2.1 is either going in the bin, or back to the supplier, unfortunately.
The one side-by-side 1.3 vs 2 photo set at night from forum at raspberrypi.org ssuggests RPiCamera v2 is just over 1 stop more sensitive than v1.3. However, every v2 photo I have seen shows misfocus problems. In the two examples on the forum, one unit is clearly focused very close (50 cm) so distant objects are blurry, and one has off-axis lens tilt (focus plane not parallel to sensor).
Well that seems something of a disappointment to me, having just taken delivery of the new cam. I ordered the NoiR version as my earlier cam is the daylight one. Apart from the obvious difference in (apparent) exposure I guess it would be easy (if you have the original images) to calculate the image area of both and from that calculate the pixel density. Looking at the overlaid image mentioned above in AndrewS’s post, and taking some measurements directly from my monitor, I’ve estimated that the pixel density of the old camera is about 205pixel/mm2 and the new is 208pixel/mm2, so resolution is pretty much unchanged. Can anyone verify these figures?
I “upgraded” an existing camera module in one of my MJPEG webcams and found the 2.0 camera board has very poor focus and the Automatic White Balance is defective, in dark scenes not enough light is captured, in bright scenes sunlit sky. Side by side comparisons of two otherwise identical webcams showed performance was worse in dark, well lit, and bright scenes. I’m returning mine as defective and stocking up on the o0ld ones.
Thanks for that! Makes for a more informed choice for a PiCam.
Does the v2 Camera have same red shiny on the center of the image? like the 1.3?? (known bug of the chip)
Any chance we could get an update with the infinity focus fixed PiCam2s? I believe the license plate test will come out better. And I’m curious if any other updates have changed the darkness issue.
I don’t have one of the latter ones yet
Please, if anybody knows: what is the diameter of the new v2.1 8mp lens; 6mm or 7mm? (m6 or m7 mount?) Thanks!!
Can anyone help me regarding choice of camera between NOIR v1.3 and V2.1. I am working on a project “Driver vigilance detection system in car”. Which camera provides me with good results as i am going to perform image or video processing?
Either should be fine. I’m told the focus issues of the early 2.1 were addressed very quickly at the manufacturers.
Think about which definition you’ll use to do your image analysis, the higher resolution, the more CPU/Time you’ll need.
what is the field of view of the v1.3?