Wide Dynamic Range Myth

[WJM]

Oh, forgot another approach/application: time-lapse compression, on longer trips.
(I know some late-generation dashcams now have this option too, more than just the Mobius 1080 that started it (I think))
(but I want that in HDR-Art form too, of course....)

(you could do that with a regular P&S, and either internal or external trigger, but then you end up with a sequence of JPEG's, which offers much less compression than MPEG's)

 

[Nigel]

Okay, now it starts getting interesting, with more facts than I had at hand....)

Second, your sequence of 1/35 - 350 - 3500 is a real (possible) sample?

I did say "something like" before my sequence, I don't know the exact exposure times or even number that are used and obviously it changes depending on lighting conditions.

As far as I can tell by analysing the results, it does use 3 exposures at least most of the time, you can sometimes see the separate exposures when there is very fast motion somewhere in the image. The longest exposure can't be slower than about 1/35th due to the frame rate and I doubt the fastest would be much faster than 1/5000th as I don't expect it is capable of reading the data out of the sensor faster than that and the data isn't buffered, it is read out a line at a time with the "rolling shutter". As I said, it doesn't quite cope with the brightness of HID headlamps so it obviously isn't very fast.

I'm not sure there is any need for more than three images since that allows you to cover the full range you are likely to get in as much detail as can be stored in a normal video file.

First, regular HDR, be it video or stills, shouldn't look artsy all, at least not by itself; the 'Art' factor is based on rather wilful tweaking (color mapping).
(IIRC, without any tweaking at all, the combined images will look dull/flat)

I have an Olympus that produces HDR images that look a bit flat and unnatural, it's an understandable effect since you have effectively turned down the contrast in order to fit the wide dynamic range into the narrow dynamic range of the video file, the mini 0806 somehow manages to keep it looking natural even though when you think about the image it is obvious that the contrast in the parts of the image in the shadows must be very different to that in parts of the image in the sunshine.

 

[Nigel]

Oh, forgot another approach/application: time-lapse compression, on longer trips.
(I know some late-generation dashcams now have this option too, more than just the Mobius 1080 that started it (I think))
(but I want that in HDR-Art form too, of course....)

(you could do that with a regular P&S, and either internal or external trigger, but then you end up with a sequence of JPEG's, which offers much less compression than MPEG's)

What is wrong with just speeding the video up? That way you can also have the sound

My 0806 showing WDR and 8x normal speed, and suffering from YouTube low bitrates! :

 

[WJM]

What is wrong with just speeding the video up? That way you can also have the sound

My 0806 showing WDR and 8x normal speed, and suffering from YouTube low bitrates! :

Hmm....have difficulty imagening what compressed audio should sound like....)
(somehow it feels not completely in sync?....)

But the advantage of time-lapse is foremost in the data volume, especially on long-distance trips....slicing the data from 30fps to 1fps means a factor 30 less data (MPEG-compression geeks might correct me on that exact calculation....)
(as there is less similarity between frames that lay 1s apart than 1/30s)

Of course, just as with my complimentary HDR-Art-Video cam, you would put a regular (speed) dashcam next to the time-lapser.
(not sure how good a time-lapse video would be as proof against speeding and/or other traffic offenses....much better than nothing, I would guess, especially for lack of any current decent automatic data-transfer, from dashcam to big file storage)

(also guessing/hoping that the GPS-data channel keeps working in time-lapse mode?)

 

[Nigel]

Hmm....have difficulty imagening what compressed audio should sound like....)
(somehow it feels not completely in sync?....)

Seems like maybe 1/10th second delay on the audio after hitting the cattle grids and water splash on my computer - speed of sound delay? Maybe your listening via bluetooth?

Are you suggesting that it shouldn't sound like that? Maybe you are thinking that I was driving fast and so it should have been high revs speeded up to super high revs when I was actually rarely going much over 2000 rpm, never went above the speed limit or there would have been a warning sound.

But the advantage of time-lapse is foremost in the data volume, especially on long-distance trips....slicing the data from 30fps to 1fps means a factor 30 less data (MPEG-compression geeks might correct me on that exact calculation....)
(as there is less similarity between frames that lay 1s apart than 1/30s)

Of course, just as with my complimentary HDR-Art-Video cam, you would put a regular (speed) dashcam next to the time-lapser.
(not sure how good a time-lapse video would be as proof against speeding and/or other traffic offenses....much better than nothing, I would guess, especially for lack of any current decent automatic data-transfer, from dashcam to big file storage)

That was only 8 times, 30 times is much too fast for replay, and only 1 frame per second original speed means you miss a huge amount, sometimes an accident happens in under a second.

(also guessing/hoping that the GPS-data channel keeps working in time-lapse mode?)

Don't know of any dashcams that do time-lapse and also have a GPS logger...

 

[WJM]

Seems like maybe 1/10th second delay on the audio after hitting the cattle grids and water splash on my computer - speed of sound delay? Maybe your listening via bluetooth?

Are you suggesting that it shouldn't sound like that? Maybe you are thinking that I was driving fast and so it should have been high revs speeded up to super high revs when I was actually rarely going much over 2000 rpm, never went above the speed limit or there would have been a warning sound.

Hmm....but if your video speeds up by a factor 8, your audio should too, meaning it would sound like 2000x8=16000rpm....)

That was only 8 times, 30 times is much too fast for replay, and only 1 frame per second original speed means you miss a huge amount, sometimes an accident happens in under a second.

Well, if the trip is long, like days/weeks/months on the road, 1/30 becomes a b(e)arable minimum.
Like one hour video for every 24 hours driving, or say 1 hour each 3 days.

Really wish there had been such camera's 10 years ago, when re-enacting a certain famous trip from 1907....;((
Would still have gained 10-20 hours video total....)
(about 3 months start-to-finish, including rest-days, or 2 months driving)

Ah, of course we would actually need some kind of dynamic frame-rate, speeding up when things become interesting, either controlled by hand (voice!....), or ground-speed related (I am pretty sure this is an option for some camera's/concepts already (of course those having GPS-log in the first place)).
(but a steady frame rate would show the ground-speed differences better)

Don't know of any dashcams that do time-lapse and also have a GPS logger...

Wasn't time-lapse (surprisingly) added to one (or all?) of the latest high-end DOD's?
(still not GPS-triggered frame rates....I think)
(like having a setting of '1 frame pro 100m')


Btw, G-sensor triggered crash-loop could also be replaced with a voice-trigger....each scream in terror means a write-protected loop....)

 

[Nigel]

Hmm....but if your video speeds up by a factor 8, your audio should too, meaning it would sound like 2000x8=16000rpm....)

Correct, and there are 60 seconds in a minute: 16,000 / 60 = 266 Hz
and there are two pistons firing per revolution on my 4 stroke 4 cylinder engine = 533Hz primary exhaust note at 8x play speed instead of 66Hz at 1x play speed.

I think what is confusing you is that you don't know what my engine sounds like in real life and a lot of the sound you can hear in the video started off at too lower frequency for the human ear to properly hear

 

[Stev]

I know that niko has recently created some posts regarding how many cameras who claim to have Wide Dynamic Range, actually do not have it. This is specific to cameras with the Novatek NT96650 Processor and Aptina AR0330 CMOS sensor which have become very popular this year. I have done some research on this topic as well and he is correct. I have updated the FAQs on these cameras with the following info:

Dynamic Range is measured in decibels (dB) and most cameras have a dynamic range of 60 dB to 80 dB. The human eye has a dynamic range of approximately 100 dB. Anything higher than 100 dB is often referred to as "Wide Dynamic Range" (to be confirmed).

While the Novatek NT96650 Processor supports Wide Dynamic Range, the CMOS sensor used, the Aptina AR0330, has a dynamic range of 72.4 dB (see datasheet). Hence from a pure technical perspective, this camera is not providing true Wide Dynamic Range. Note that a high dynamic range number (>100 dB) does not guarantee great night performance if other aspects of the camera are poor (processor, lens, firmware).

There is a lot of technical information behind this which I'm sure niko or others can elaborate on. I am not an expert in this area by any means so I am open to suggestions on this topic. Also note that even if a CMOS sensor claims a high dynamic range number on a spec sheet, the actual measured number might be less and apparently there are standard ways to measure this.

I'm interested if these sensors have dedicated HDR/WDR modes? The normal dynamic range quoted is the native dynamic range, but if the chip has HDR etc video mode, it can add a number of stops on top of native. I had old cameras with maybe 8+ stop native, but upto 16 stop+ in HDR. But word of warning, the more added stops the worse the quality in some of these HDR schemes.

So, if this camera had 11-12 stops native, then quality HDR might be a few more stops (which might be what we are seeing).

 

[Stev]

Actually. I'm looking around for the camera that claimed something like 20 stops. Anybody know of any?

 

[Nigel]

Actually. I'm looking around for the camera that claimed something like 20 stops. Anybody know of any?

Why?

Our dashcams store their images in H264 video files, and sometimes in jpg photo files, both of which contain 8 bit colour. How are you going to store your 20 stops inside an 8 bit file?

 

[Stev]

Bits are not the same as stops. The bit just represents a quantity. Say photons are marbles, and the bit is a can. The can hold one or more marbeles. So, a bit could reptesent one or more photons. It could represent all the photons in the universe even. The bits merely tell how many times the minimum quantity doubles (unless you use logarithmic). The simplest way is to align the bottom bit with the desired signal floor and the top bit the desired ceiling, making the bits represent the division of the signal inbetween.

They already record more than 8 stops in 8 bits on cameras here.

 

[Nigel]

Bits are not the same as stops. The bit just represents a quantity. Say photons are marbles, and the bit is a can. The can hold one or more marbeles. So, a bit could reptesent one or more photons. It could represent all the photons in the universe even. The bits merely tell how many times the minimum quantity doubles (unless you use logarithmic). The simplest way is to align the bottom bit with the desired signal floor and the top bit the desired ceiling, making the bits represent the division of the signal inbetween.

They already record more than 8 stops in 8 bits on cameras here.

Yes, each bit represents a power of two - a doubling of the value,
and each stop also represents a power of two - a doubling of the amount of light,
so a sensor capable of 20 stops will need to output 20 bit data values to the processor.

Our dashcams don't encode their 8 bit data values in H264 and jpg linearly, they use ITU-R BT.709 for video and sRGB for photos, and are able to encode more than 8 stops of dynamic range, but 20 stops?

Remember that 8 bits means that there are only 256 possible values, 20 bits means 1,048,576 possible values, so you are going to have to loose 1,048,320 possible values before it will fit, that is a lot of lost detail! And having done that you can then set about the H264 compression and loose even more detail!

 

[Deleted member 24437]

Western innovation?