Timelapse with variable shutter speed

fbonomi

I was doing some logging of BV changing in time with the program above, and I found out something interesting.

The following is a graph of BV taken with over 800 samples tonight from 19:15 up to 21:30, covering the full sunset.

You can see the luminance decreasing gracefully down to -200, then you hit the sensitivity of the CCD.

To put it briefly, while you may be able to shoot in those very dark conditions, you won't be able to reliably calculate expositions beforehand.

fudgey

How about calculating exposure from a test shot?

This would of course require new functionality in CHDK, but the idea would be to shoot first, then calculate exposure from the resulting RAW image in RAM, delete this test shot JPEG (or prevent it from being written entirely if possible), adjust exposure and do the final shot (or if still off, iterate until a desired exposure has been achieved).

fbonomi

Yes, this is certainly an option.

But shooting (for example) a 60-seconds shot just to measure luminance would make my shots too far away in time. (This is already a problem, when the exposure grows!)

In this case, I am helped by the fact that I am shooting a series of shots that are rather close-up in time.

I could therefore:

Activate preview mode
Take shot N of my sequence
Being shot with a long exposure, I will have on the LCD a image much brighter than "live preview"
Measure "luminance" of preview
See difference with image N-1
Extrapolate this difference to adjust exposure for image N+1

etc.

But for the moment I might take a simpler approach: if I am shooting a sunset, after the Bv goes below -200 go in a sort of "guess mode" and simply assume that Bv is decreasing at a certain, constant rate...

After all, I don't want scientific results, I just want a nice timelapse. This would mean that at a certain time in the night (something like 21:00) there is a sort of slow "fade in" bringing the sky to the maximum brightness possible.

Of cource, this would NOT work for a dawn 😀

wontolla

Nice! From 1000 to -200 we couldn't ask for a smoother curve.

Another suggestion: What if you increase ISO when reaching the -200 problematic area?

That way, the camera could be able to still calculate exposure accurately and also you can keep a fast shutter speed. With the drawback of noise of course, but if you plan to upload your time lapses (high compression and low resolution) noise could be no problem.

But the extrapolation thing sounds good. Even without that calculation, you can "hard code" a slope around 45 degrees in your graph.

I'll try to run your test script tomorrow. I was planning today but had to stay late at work.

fbonomi

wontolla
Another suggestion: What if you increase ISO when reaching the -200 problematic area?

Well, I am not strong in this area, but let me try to explain what I think I grasped 😀

We are here speaking of Bv, i.e. of luminance.

This is completely independent of ISO, this is just saying "how much light is coming in?"

This is the "luminance" value (BV): the camera feels how much light is coming to the CCD and measures it.

If we see steps in the graph is because we are here meeting the physical sensitivity limits of the CCD

This is the start of the data flow: beginning with this value, the camera starts to virtually emulate the exposure in a film camera.

To change sensitivity in a film camera, one would replace the film with one with a different ISO.

In a digital, we can't change CCD and them we emulate ISO, by decreasing the numbers read from the CCD

THEY CAN ONLY BE DECREASED, NOT INCREASED: the CCD sensitivity stays the same, and if the light is less, the CCD just doesn't "feel" it!

With ISO, we have Sv. Then, we look at the selected aperture (Av), i.e. "how much will the iris close when we shoot?"

We are now able to apply the APEX formula

Av+Tv = Bv+Sv

we have Bv, Sv and Av, and then we can transorm this into:

Tv = Bv+Sv-Av

Bingo! we are able to calculate Tv, that gives us the correct exposure time...

fbonomi

Ok, first results. 930 frames, from 19:17 to 23:54, with expositions ranging from 1/2000 to 60", ISO 200, shot with A570 IS

Shot opposite to the sun, so no purple colors. At the beginning, (a few seconds in the movie), you can see a strong light in the hills, it's the sun reflecting in a large window of a house!

You can see the results here:

Sunset time-Lapse n. 4

If at all possible, look at the VERY large (22MB) original file here:

Sunset time-Lapse n. 4

Where you can see a few stars (practically invisible inthe flash version).

Again, good news and bad news.

The good news is that the thing seems to work, and once the problems are sorted the results might be impressive.

Also, in a clear sunset the light curve is very smooth, so a less sofisticated algorithm might be enough. It's all another matter if you have pathes of clouds, the sun appearing and disappearing between the clouds makes it very difficoult to compromise between a "smooth" curve and a faithful one.

This was shot opposite to the sun, if you shoot towards the sun the curve will become much harsher (don't shoot directly to the sun!)

The bad news:

1) BY MISTALE I left the script running with a debugging value (minimum step = Tv 16) so you can still see "bump" il luminosity (they are the steps in the Tv graph below). You can se these "flashes" in the movie, and as the Tv curve gets steeper and steeper, you can see the flashese get faster and faster) See the graph below. When Bv gets under -200, the step becomes 5 Tv (the script goes in "guess mode", see below) and you see that the steps alost disappear. That's why, in the last seconds of the film, the fade-in is much smoother

2) The exposition gets longer and longer. So, with a delay of 3 seconds between each shot (like in this case) you have the following:

At the beginning the movie shoots for 1/2000, then waits for 3".
At the end it shoots for 60 seconds, hangs about 30 seconds for image post-processing, then waits another 3 seconds

So, at the beginning we have a shot every 3 seconds, at the end we have a shot every 93 seconds. (the numbers aren't real, but you get the point)

This means that the playback time is greatly compressed in the "night" portion. The firs half of the movie gets about 850 frames, the second half gets about 80 frames.

The middle part, whee the sky gets progressively darker and darker, is boring in the movie because the sky gets darker, the camera exposes more and it seems nothing is changing 😀

I should have started with a larger delay to start with, but the problem stays.
Probably the solution would be to increase exposure time up to a certain point (let's say 15 seconds) and then start increasing Sv, not Tv (increasing ISO, not lengthening shutter time)

3) Stars almost completely disappear when compressing the movie. They are wonderful in the uncompressed fom, but that's 1.4 Gb 😀
Some post-processing (or compressing) trick must be studied, but probably this problem is less severe with other subjects (e.g. the lights of a town)

4) the script goes in "guess mode" when Tv < -200 (see post above) so the script in its current form is good for sunsets but not for sunrising. Will have to wake up early to get a Bv curve for a sunrising. I guess for all practical purposes it's lke having a sunset reversed, but I want to see the Bv curve in the very dark portion, to see how the large "steps" below -200 appear and try to cope with them

Ok, I will give it another try tonight and then release the script (it's yet in a very unpolished form, with no parameters etc.)

Thanks to the ones that gave me their parameters, they have helped a lot!

dkeggins

fbonomi

First, thanks for the script and look foward to any improvments. I am new at this (not photography) but scripts to perform with. I have the S5IS and I had to make one modification to the script to make it work. I had to change the 'press "shoot_half"' lines to shoot_full. With it set to shoot half, it would only take the first frame. It would continue the log from that point on but wouldn't actually take any pictures. Again, thanks and I hope to learn more on the scripts and make some of my own.

fbonomi

the script was only supposed to take one shot every 60 measurements: measure every 10 seconds, shoot every 10 minutes.

This because I was just using the script to collect data, not actually shooting, and I needed the shots just to get some clue of what was happennig in the meanwhile.

the press "shoot_half" is there just to (supposedly) help the camera in measuring.
the place where the script actually shoots is this:

if (p%60=0) then shoot

which means: if p (the measurement counter) is divisible by 60 then shoot.

so the script will shoot at measurement 0, measurement 60, measurement 120 and so on.

if you want it tio shoot alway, just replace the whole line with just:

if (p%60=0) then shoot

but mind you, this script does not yet adjust exposure... it's just meant to collect data!

fbonomi

for the daring ones....

a first, untested version of the script. It's UNTESTED, I will try it tonight, if I succeed.

The idea is:
1) forget (for the moment) the smoothing. That will be added later
2) the script starts in standard mode: ISO 100, Tv calculated
3) when it gets too dark and (we risk getting bad measurements from the CCD) we go in guess mode: the Tv is increased of 5 every shot.
4) this goes on up to a Tv equivalent to about 20 seconds: when exposure would get even darker, we go in "guess mode with high ISO": this means we keep exposure fixed to 20 seconds and then we start increasing ISO instead.
5) this goes on up to about ISO 800, after this the camera keeps shooting with the same parameters.

Please note that there are some odd behaviours in the get/set_tv96/sv96 commands, so if you try the script by hand and make VERY fast changes (like going from outdoor to indoor ) you might get one or two overexposed frames.

All this said, here is the script. give it a try, if you want to!

EDIT: the code below is buggy... will have to study better

@title Sunset Time Lapse

print_screen 1

print "Sunset Time Lapse"


rem minimum Tv (-414=20 sec)
M=-414

rem minimum Sv (-776=800 ISO)
N=776


rem Guess mode limit
G=-200

rem initialize guess mode value
x=G

rem get start Tv
get_tv96 T

rem get default Sv
rem get_sv96 S
S=5*96-1
print "Default Sv: ",S

rem picture counter 
p=1


:loop

rem measure luminance and aperture
press "shoot_half"
sleep 500
release "shoot_half"
get_bv96 B
get_av96 A
rem release "shoot_half"

print "Measured: ",B ,A

rem resulting Tv would be:
T=B-A
T=T+S

print "Calculated T: ", T

rem check Tv Values

if T<-200 then
 rem normal mode, shoot with calculated Tv and default Sc
 X=T
 Y=S
 print "Mode: Standard"
else
 rem Guess mode, every shot will be 5/96th steps longer
 x=x-5
 

 if  x>M then 
  rem we are in dark, but exposure is not too long
  rem normal mode, shoot with guessed Tv and default Sc
  X=T
  Y=S
  print "Mode: Guess"
 else
  rem we are in very dark area, we want to avoid too long exposures
  rem shoot with maximum allowed time (M) and
  rem adjust Sv accordingly

  X=M
  Y=S+M-x
  
  rem BUT check we don't go in too high ISO!
  if Y>N then
   Y=N
  endif
  
  print "Mode: Guess with High ISO" 
 endif
 
endif


rem we can now shoot
set_tv96_direct X
set_sv96 Y
print "SHOOT ",p,X,Y
shoot

p=p+1

sleep 2000

goto "loop"

fbonomi

It turns out that the script is calculating properly exposure and sensitivity but the shots are not taken with the specified Tv settings (some have the correct exposure, but a lot have 1/1250).

Below is the graph, derived from its logs, of the script's calculations (black is Tv, pink is Sv).

990 shots, you see the gradual change in Tv (unsmoothed, so there are some ripples), then (at the very right) Tv gets to -200 and the script goes in guess mode (slope is now much more uniform.
At -414 Tv starts decreasing and Sv increases.

The problem is, the script was trying to set Tv, but not always this has happened, especially at long exposures

fbonomi

Here is the movie, assembled up to the point where the camera started to get wrong frames (clik on image)

It's absolutely not bad.

The exposure is not smoothed, so you can see some ups-and-downs in brightness, otherwise, the camera compensates almost perfectly for the increasing darkness, with exposures varying from 1/2000" to 2.5"

The very last part is nice, with the lights of Florence turning on in the valley.

It's a shame it's so short, but that is when the camera starts getting it wrong 🙁

wontolla

Indeed, the last part is very good, the blue colour of the sky and the lights of the city. Well done.

So, fbonomi, can we use the script you posted? Is it still buggy?

Continue with this good work!

fbonomi

yes it is!

I think I have understood the problem, I will try it again tonight and post it.

I can only test it properly once every 24 hours (at best)

😀

fbonomi

Another version of the script.

It is now working, but there is still much space for improvements

The logic is the following:

1) Standard Mode: we start with a default ISO setting (Default Sv, default 100 ISO) and the camera shoots changing Tv accordingly.
2) Guess Mode: when the scene is too dark (Below Guess mode Limit, default -200, this means that the camera would need to expose more than 4 seconds) we probably cant' rely any more on reading the CCd on the fly, therefore we just keep exposing more and more every shot (Tv is increased by "Slope in guess mode", default 5)
3) Guess mode with high sensitivity: when exposure gets too long (limit Tv, default 20 secs) the camera does not make longer exposure, but rather increases sensibility, going up to "Limit Sv" (default 776, ISO 800)

The program logs quite some information in the PR_SCREEN.TXT file (in CHDK/SCRIPTS)

All values (in the log file and in the parameters) are in the APEX*96 format

Feel free to test this script, as always it would be very useful for me to receive the log file

@title Sunset3

@param a Delay (sec)
@default a 5

@param b Limit Tv 
@default b -414
rem (-414=20 sec; -384=15 sec; -320=10 sec; -224=5 sec.)

@param c Default Sv
@default c 480
rem (480=100)

@param d Limit Sv 
@default d 776
rem (776=800 ISO; 960=3200)

@param e Guess mode limit
@default e -200

@param f Slope in guess mode
@default f 5


print_screen 1

print "Sunset Time Lapse"

rem initialize guess mode value
x=e

rem get start Tv
get_tv96 T


rem picture counter 
p=1


:loop

rem measure luminance and aperture
press "shoot_half"
sleep 500
release "shoot_half"
get_bv96 B
get_av96 A
rem release "shoot_half"

print "Measured: ",B ,A

rem resulting Tv would be:
T=B-A
T=T+c

print "Calculated T: ", T

rem check Tv Values

if T>e then
 rem normal mode, shoot with calculated Tv and default Sc
 X=T
 Y=c
 print "Mode: Standard"
else
 rem Guess mode, every shot will be 5/96th steps longer
 x=x-f
 
 if  x>b then 
  rem we are in dark, but exposure is not too long
  rem normal mode, shoot with guessed Tv and default Sc
  X=x
  Y=c
  print "Mode: Guess"
 else
  rem we are in very dark area, we want to avoid too long exposures
  rem shoot with maximum allowed time (b) and
  rem adjust Sv accordingly

  X=b
  Y=c+b-x
  
  rem BUT check we don't go in too high ISO!
  if Y>d then
   Y=d
  endif
  
  print "Mode: Guess with High ISO" 
 endif
 
endif


rem we can now shoot
sleep 100
set_sv96 Y
set_tv96_direct X
sleep 100
shoot
sleep 100
print "SHOOT ",p,X,Y
p=p+1

sleep a*1000

goto "loop"

Know Issues (things that might be changed)
1) The calculation is not smoothed, therefore the exposure occasionally bumps up and down a little (I will insert a smoothing algorithm, but before I want to be sure the camera exposes correclty
2) below a certain level of light, the camera just keeps exposing more and more up to the specified maximum. This can lead to over-exposure. I probably could trust a bit more the CCD (i.e. consider its values even for darker regions) and/or find a way to measure lumnance not from the CCD (on the fly) but based on previous photo (which has probably been exposed a lot)
Otherwise, if one knew in advance the corect maximum exposure, one could for example lower the maximum allowed sensibility..

fbonomi

The experimental movie of tonight....

It was overcast, and not a terrific location, so not quite as impressive. Interesting anyway, and anyway Ineeded to test the script 😀..

630 frames, from 19:30 (ISO 100, 1/1250) to 23:20 (ISO 1250, 15 sec) Gamma 1.6 applied to the movie

The final part is quite overexposed. Also, the "Slope in Guess Mode" was set to 20, and this caues a too fast transition from the "dark blue" sky to the "orange" sky.

The last part being over-exposed, one coyuld set a shorter maximum exposure and would also obtain the result of "slowing down" the clouds in the dark area

The following is a graph of the parameters (Measured Bv, Calculated Tv and Sv)

fbonomi

Well, I think we are there!

My movie of last night:

800 frames taken from 19:10 to 00:47, starting from ISO 100, 1/1000" and ending to ISO 1250, 8".

It seemed that it was going to be overcast, but a while after sunset it got clear and the results are excellent.

Some post-processing was needed, as I underexposed the last part of the video.

This is a limit of this script: the final part of the video (the guess mode) is not auto-exposing, you have to set manually the final Tv and Sv, and it's not always easy to guess how it will be in a few hours. Anyway. that's unavoidable without changing the CHDK source itself, and therefore will be left to v. 2.0 of the script.

fbonomi

FINAL VERSION OF THE SCRIPT
(and instructions)

This script attempts is meant to shoot time-lapse videos of sunsets: starting from full daylight and ending in full night, always adapting exposure for a correct shot.

Up to a certain point (when there is enough light to be directly measured) the script exposes automatically, then starts guessing the exposure, increasing first exposure time (up to a certain limit) then ISO (again, up to a certain limit).

Suggested camera settings:
Camera mode: manual
White balance: Sun (other settings, if desired but not auto)
Image size: a small image size (S) is suggested, in order to allow more shots
Flash: off
Focus: manual, set to infinity
Aperture: maximum

Note: all the parameters here are expressed in APEX*96 units, with equivalence.

First, you wll have to decide the delay between one shot and the other (how many seconds between each shot)
Suggested values: 5 or 10 seconds. Longer might also be ok, bu the resulting video will be shorter.

Then, you will have to decide what will be the final exposure parameters. These will depend strongly on the type of scene you will be shooting.

Examples of suggested values:
Landscape with strong lights (city), illuminated sky 5" (Tv -224) ISO 1250 (Sv 767). Stars will be barely visible
Landscape without lights (countryside), dark sky 15" (Tv -384) ISO 2500 (Sv 864) Stars will be well visible

These parameters are called "Limit Tv" and "Limit Sv" in the script.

Finally, you have to decide the "Default Sv", (the ISO settings to start with). I suggest ISO 100 (Sv 419)

Other values ("Guess mode limit" and "Slope in guess mode") are better left to their default values

You can now run the script.

For debugging purposes, the program logs quite some information in the PR_SCREEN.TXT file (in CHDK/SCRIPTS). If you have problems with the script, save the PR_SCREEN.TXT so that problems are better pinpointed.

This is the script:

@title Sunset4
rem Script to shoot time-lapse videos of sunsets
rem v. 4, Fbonomi apr 30th 2008
rem Released under GPL

@param a Delay (sec)
@default a 10

@param b Limit Tv 
@default b -414
rem (-414=20 sec; -384=15 sec; -320=10 sec; -224=5 sec.)

@param c Default Sv
@default c 480
rem (480=160)

@param d Limit Sv 
@default d 776
rem (776=1250 ISO; 960=5000)

@param e Guess mode limit
@default e -200

@param f Slope in guess mode
@default f 5


print_screen 1

print "Sunset Time Lapse"

rem initialize guess mode value
x=e

rem get start Tv
get_tv96 T


rem picture counter 
p=1


:loop

rem measure luminance and aperture
press "shoot_half"
sleep 500
release "shoot_half"
get_bv96 B
get_av96 A
rem release "shoot_half"

print "Measured: ",B ,A

rem resulting Tv would be:
T=B-A
T=T+c

print "Calculated T: ", T

rem check Tv Values

if T>e then
 rem normal mode, shoot with calculated Tv and default Sc
 X=T
 Y=c
 print "Mode: Standard"
else
 rem Guess mode, every shot will be 5/96th steps longer
 x=x-f
 
 if  x>b then 
  rem we are in dark, but exposure is not too long
  rem normal mode, shoot with guessed Tv and default Sc
  X=x
  Y=c
  print "Mode: Guess"
 else
  rem we are in very dark area, we want to avoid too long exposures
  rem shoot with maximum allowed time (b) and
  rem adjust Sv accordingly

  X=b
  Y=c+b-x
  
  rem BUT check we don't go in too high ISO!
  if Y>d then
   Y=d
  endif
  
  print "Mode: Guess with High ISO" 
 endif
 
endif


rem we can now shoot
sleep 100
set_sv96 Y
set_tv96_direct X
sleep 100
shoot
sleep 100
print "SHOOT ",p,X,Y
p=p+1

sleep a*1000

goto "loop"

avatorl

Thanks, it's working fine http://www.youtube.com/watch?v=_xuI9_IbaqM
As I understand this is the only script for sunset timelapse? What about the sunrise modification?

fbonomi

Hi avatorl,

thanks for sharing! Where is it?

Yes, in this form it's only good for sunsets (light to dark).

I am working on a more generic script, but that will require changes to the CHDK source (the C code, I mean)

PhyrePhoX

could you elaborate on these changes?

« Previous Page Next Page »