A look shutter-induced motion of the whole camera

Started Dec 14, 2013 | Discussions thread
Flat view
Ken Strain Regular Member • Posts: 448
A look shutter-induced motion of the whole camera

There has been a lot of discussion of "shutter shock" and I was curious to see if anything could be learned by combining photographic tests with external vibrometry. The circumstances that I want to explore here here are those of hand-held photography where IBIS or OIS are used to reduce camera shake and allow shutter speeds of around 1/60s. I have seen little or no evidence that either IS type makes blur worse due to camera vibration with any lens-body combination which I have investigated.

Measurements of body (external) vibration of various Micro-Four-Thirds (MFT) cameras.

Disclaimer: It should be clear from the start that only limited conclusions can be drawn from measurements of the bulk motion of cameras in one degree of freedom at a time, and that I do not expect these results to be very useful to most photographers.

There are two prevailing hypotheses for the cause of shutter-motion induced blur in MTF cameras:

  1. the "shock" of the shutter blinds accelerating shakes internal components of the body or lens leading to motion of the image relative to the sensor during exposure
  2. the shock causes the whole camera to pitch (in landscape orientation) causing the camera to point in different directions at different times during the exposure.

I expect that both effects contribute.  Hypothesis 1 is very difficult to investigate as the internal parts are not easily accessible to vibrometry at least with a lens on. I think, however, that the other cause is perhaps more important in many circumstances.

Anders W has noted (see MFT forum) that in his experience a solid support concrete floor can reduce blur. Since the floor probably did not have resonances that coupled to the innards of the camera that gives some support to Hypothesis 2. Personal experience with a very vibration-prone (I assume faulty) E-PL2 and advice from Anders conviced me that subtle changes in grip can make an important difference to the resulting blur at least for some camera/lens combinations. Since my hands do not have high-Q internal resonances with which to couple this also points to the importance of bulk motion of the whole camera rather than internal vibration. On the other hand it seems likely that coupling of internal resonances with high-Q resonances of a tripod, could easily lead to enhancement of blur from motion within a rigidly supported body and in those circumstances Hypothesis 1 may be expected dominate. This post is about hand-held photography, where Hypothesis 2 seems to take precedence.

Tools and techniques - common features of the measurements

Apparatus: PZT shear accelerometers; a low-noise high impedance pre-amp; microphone input of a PC (calibrated), and MATLAB for sound recording (44kHz sampling) and analysis. One challenge with this type of vibrometry is in reliably attaching the sensor to the device under test to ensure fidelity at higher frequencies. To minimise damage, the accelerometers were mounted using double-sided tape. A sensitive (B&K4368) sensor was used for most measurements, and a model 4374 (around 20 times less massive) was used to check fidelity at high frequency. The measurement band was chosen after these initial tests to be from 100Hz to >3kHz. Two 100Hz low pass filters were applied (one in hardware and one in software) to yield flat response to displacement over the measurement band - with, in displacement terms, two-pole high pass for signals below 100Hz.

Direct measurements of pitch were not achieved, but some measurements of motion at the end of a longer lens show that pitch is an important component of the motion (as may be expected when a compound mass receives impulses far from the centre of mass).

5s stretches of data were taken with 3 or 4 shutter actuations per stretch. These were all hand held. Consistency was good in all cases, and the single actuations that are shown are representative.

After the low pass filtering, analysis consisted of multiplication by a calibration constant believed accurate at the ~20% level, although there could be systematics that have been missed. Three types of plot were produced. Time domain displacement was plotted without further processing and is thought to have an accurate scale. Spectrograms were generated using the PSD result from the MATLAB spectrogram function (plotted here as the A/PSD in dB rel. 1 m in 1 Hz). It is the shape of the spectra that matter, and I have chosen a scale in which colours yellow through red indicate progressively greater chance that the vibration would lead to blurring. I am uncertain about the precise scaling. Finally each time-stripe from the spectrogram was integrated over frequency and the result plotted logarithmically to potentially reveal "ring-down" of dominant resonances (the units are not really arbitrary, but I have not checked all the factors required to scale back to displacement).

Shutter induced vibration has the greatest effect in these MFT cameras in the range from perhaps 1/30s to 1/125s. To help identify the times at which the shutter was open, 1/10s exposures were also made.  That means that there were only a few hundred samples during the exposure time, or a few cycles of the dominant frequencies.  Time domain results are quite informative, but the resolution of the spectrogram is limited.  Only a small subset of the data are shown here (but without any selection based on seeing the results).


The cameras available were E-PM1 (its IR filter was replaced with a clear glass), two E-PM2s and a GX7 which has the option of an electronic shutter. Two lenses were used: 45/1.8 with small moment of inertia and the larger 35-100mm. A post in the MFT forum shows some photographic tests for blur in these cameras with these lenses.

The plan was to make measurements to identify which part of the shutter cycle corresponded to the exposure, to see how consistent the vibration was, to explore the effect of how the camera was gripped, to look at motion at the end of the lens as well as on the body, and to compare different cameras of the same and different types.

The results have been obtained and are presented separate posts.

In preparation for the results it is worth considering what level of vibration matters. The threshold for seeing blur is hard to determine since the blur function is not really known. Some artificial test images  suggest that if sinusoidal motion of 1 pixel rms occurs over 1/3 of the exposure that would be visible, but much less motion or smaller duration would probably not be. Taking a relatively short focal length for which blur seems to be a common problem (45mm) the angle of view equivalent to 3460 pixels is 17 degrees, so around 80 microrad per pixel. With about 45mm from sensor to entrance pupil (for the 45/1.8 lens), that suggests that the range of interest is vibrations of amplitude of order a few microns lasting of order 1/3 of the exposure.

To be continued ...


Olympus PEN E-PL2 Olympus PEN E-PM1 Olympus PEN E-PM2 Panasonic Lumix DMC-GX7
If you believe there are incorrect tags, please send us this post using our feedback form.
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