Timelapse with CHDK and Avisynth

A timelapse is an animation technique that consists in shooting frames at a certain speed and playing them back much faster, giving a greatly accelerated video effect. This allows to see happen in a few seconds phenomenon that normaly take hours, days, or months. It's great fun to watch, and with a little equipment and some patience, it can be quite easy to make your own.

I wanted to use only my point and shoot camera to take the shoots and free software to do the processing on the computer. I used a Canon Digital IXUS 750 (also known as SD550) camera with the CHDK firmware.

You need a camera capable of taking shots at periodic, programmable intervals to take the pictures. Most newer, higher-end cameras have the feature built-in, but you can do it too with a low-end Canon camera provided you install the CHDK firmware. It doesn't actually modify the firmware of the camera, so there is no risk to "brick" your camera and you can go back to the original firmware at any time.

A great feature of CHDK is that it can interpret BASIC or LUA script directly, which allows to completely automate the camera and add features normally only available on the higher-end models. I made a simple LUA script to take pictures at regular intervals. You can copy it below, and save it as timelapse.lua on your memory card (put it in the /CHDK/SCRIPTS/ directory). I only tested it with my Canon IXUS 750, and some features are not available, or behave differently on different cameras, so some tweaking may be in order. However for the most part it should be the same :

--[[
@title Time Lapse
@param n Number of frames
@default n 0
@param m Number of frames x100
@default m 1
@param i Interval in seconds
@default i 5
@param d Delay before start
@default d 3
@param r Refocus each frame (1=yes)
@default r 0
]]
 
totalFrames = (n > 0 and n or 0) + (m > 0 and m or 0) * 100
interval = i > 0 and i*1000 or 1000;
delay = d > 0 and d*1000 or 0;
refocus = r == 1 and true or false
 
print( "Shooting "..totalFrames.." frames every "..(interval/1000).."s for "..(totalFrames*interval/1000).."s." )
 
if delay > 0 then
  print( "Starting in "..(delay/1000).."s" );
  sleep( delay )
end
 
if not refocus then
  press( "shoot_half" )
  sleep( 2000 ) -- Allow a little time to focus
end
 
nextTick = get_tick_count() + interval
for i = 1, totalFrames do
  print( "Frame "..i.."/"..totalFrames )
  if refocus then
    press( "shoot_half" )
    sleep( 2000 )
  end
  click( "shoot_full" )
  if refocus then
    release( "shoot_half" )
  end
  sleep( nextTick - get_tick_count() )
  nextTick = nextTick + interval
end
 
if not refocus then
  release( "shoot_half" )
end

The script parameters are :

• Number of frames : Number of frames to shoot.
• Number of frames x100 : Same as above, but times 100. Use this to save some keypresses :).
• Interval in seconds : Interval between each shot.
• Delay before start (in seconds) : Wait a bit to allow you to get away before the shooting commences.
• Refocus each frame (1=yes) : If 1, will perform a full shutter press for each frame, triggering the camera autofocus, exposure and auto white balance. If 0, the focusing, exposure and white balance will only be done for the first frame, and all subsequent frames will use the same values.

Example configuration :

Take 10 000 shots (in other words, take pictures until the memory card is full, or until I decide to stop) at 2s intervals, wait 5s before start, don't refocus each frame.

You can then start the script :

You'll get a quick estimate of the total time and the shooting will start.

Once the script ends, you'll hopefuly end up with a few thousand pictures. Now you need to process them into a movie. First copy everything in a temporary folder on your hard drive.

Avistynth is a very nice piece of software with allows to process videos through text scripts. It's extremely powerful, but the particular feature we're interested in is that it can use images as a video source.

If you've got a Canon camera like me you should have a directory structure with each directory containing a couple thousand pictures. Assuming you put the pictures in "e:/tmp/lapse/images/", here's what the Avisynth script could look like :

ImageSource(file = "e:/tmp/lapse/images/123CANON/IMG_%04d.jpg", start = 2001, end = 4000, fps = 30) +\
ImageSource(file = "e:/tmp/lapse/images/124CANON/IMG_%04d.jpg", start = 4001, end = 6000, fps = 30) +\
ImageSource(file = "e:/tmp/lapse/images/125CANON/IMG_%04d.jpg", start = 6001, end = 7000, fps = 30)
 
# If you wish, you can also resize the images at this step.
LanczosResize(800, 600)
# You can crop too. Make sure the final dimensions are multiples of 8.
Crop(20,20,760,560)
 
# This last step is needed to make sure the image data is in a format friendly to the encoder.
ConvertToYV12()

Since I'm lazy I made a small Python script to create the ImageSource part of the Avisynth script for me. It's pretty much a hack and works only if the images are named IMG_1234.jpg like on my camera. I'm not quite fluent in Python either, but it works. Change the path as needed :

import os
 
base = "e:/tmp/lapse/images"
folders = sorted(os.listdir(base))
for f in range(len(folders)):
   images = sorted(os.listdir(base+"/"+folders[f]))
   start = images[0][4:8]
   end = images[-1][4:8]
   print( "ImageSource(file = \""+base+"/"+folders[f]+"/"+"IMG_%04d.jpg\", start = "+start+", end = "+end+", fps = 30)", f!=len(folders)-1 and "+\\" or "" );

Avisynth does the frame processing, but not the actual encoding. The last tool we'll need for this last step is MeGUI. It's a very complete x264 encoding suite, and is intended to work with Avisynth scripts. Launch the tool and load our script :

Choose encoding settings suitable to your source images and the final video quality. There are many presets built-in but you can make your own if you want. Next click the "Enqueue" Button, and click on the "Queue" tab :

I chose a 2-pass encoding so there are 2 jobs. Now click start to launch the jobs. A progress window will pop up and after a while the final video will be generated.

Troubleshooting : if the job ends immediately with an error check the following :

• If you cropped or resized the images, make sure the final size is a multiple of 8. This is needed by the x264 encoder
• Make sure the Avisynth scripts contains the ConvertToYV12() line

Here is a sample video I made. Number of frames : 5000 (2s interval). Speedup factor : 60 (30 fps). Location : Paris, over the Parc Montsouris. Shoot date : 2009-06-11.

You can download the video files directly below :

• Clouds.ogv 1MBps, 640x480, OGG/Theora format, 20MB
• Clouds.mp4 1MBps, 640x480, MP4/H.264 format, 20MB
• Clouds full.mp4 (Original file) 8MBps, 1600x1200, MP4/H.264 format, 159MB