Shake Reduction test with a laser pointer

Started Feb 24, 2010 | Discussions thread
Flat view
panasonicpentax Regular Member • Posts: 181
Shake Reduction test with a laser pointer

I really have a lot of difficulty believing the testing of shake reduction where the shake is provided by hand-holding the camera. Yes, it does accurately model the actual way that the camera is going to be used but it's not repeatable. In fact, it's not even possible to measure if shake did occur on the "sharp" pictures or the shutter went off at the lucky moment when the camera was stationary.

Ideally, the camera would be mounted in a gimbal and the shake provided by a mechanical means. Even this has problems: do you rotate by a specified distance or do you use a specified "shake energy"? Heavier cameras will obviously do better with constant energy but that's probably the correct approximation for hand-holding. If you use a "kick" impulse, how do you trigger the shutter at the correct time so that the shutter opens at the point of maximum velocity or if you use a constant (sinusoidal) vibration, how do you trigger the shutter at a consistent place in the cycle each time? What if you're trying to measure a P&S that adds a random delay due to synchronisation with the live-view frame rate? A 60th of a second is going to completely randomise the results.

Putting a human in the loop just makes it so likely that unconcious biases will distort the results and there's nothing you can do to see if these biases exist in your data set.

Well, I haven't built a gimbal yet but I have started to measure whether shake occured during the exposure. It's really simple: just attach a laser pointer to the camera. It must be rigidly mounted to the camera body and preferably not too heavy. Mine is very light because the batteries are in my pocket.

Please note the the following photos are just a proof of the concept. If this was a proper investigation, the test chart would be the same size in each image and the focus, lighting and white balance would be consistent. All images are uncropped so you can compare sensor movement against the size of the sensor.

So, first check that the laser is rigidly mounted - with SR OFF - shake the camera and check that the laser dot is stationary in the image. This takes a few attempts to be able to get shake to show in the picture and I couldn't do it repeatably.

Now turn SR ON and try the same again. This time the SR will try to make the test chart appear sharp and unmoving while the laser dot will appear to move in the final image. You will find that you have to intentionally shake the camera quite a lot and you will get many, many images where the shake is smaller than the laser dot.

To get this image, I had to use a 250mm lens and stood well back from the test chart. At this range, the laser pointer makes a rather large fuzzy blob, not a sharp dot.

Wow - the SR really is doing a pretty good job here. It's not a very sharp image, even by the standards of this test set. But it does show you just how far the sensor is being moved to compensate for the camera movement. We aren't looking at a couple of pixels here - more like 300 pixels.

No, the K20D can't correct for rotational motion (would really need 2 laser pointers to test this on a K7)

It also can't correct for pure translational motion at short diistances - if you just jiggle the camera sideways during the exposure you get the same result as the unstabilised photo above. But rotational movements are still corrected even at macro distances.

The really fun part is looking at the laser dot in live view mode. You can really get a feel for how much work the SR is doing for each movement of the camera. (Note that live view has a delay, which makes it seem more stabilised than it actually is.)

Using live view, you can immediately see how it works when panning the camera. My Pentax K20D smoothly tracks the panning shot and applies the SR corrections in both directions to any small vibrations it gets during the pan. However, my Panasonic TZ7 (pointy-shooty-zoomy-thing) switches to a panning mode and only applies vertical corrections - the image of the dot only moves vertically on screen when panning.

With a better laser pointer (sharper dot) and some more rigorous exploration of the parameters, I'm sure there is more to be learned about how SR really works and how little motion blur can be intentionally generated in a handheld test (or unintentionally suppressed.)

Flat view
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow