What does it take to photograph insects in flight? First, the camera needs to be able to take an exposure “at a moment’s notice”. The fastest consumer cameras have shutter lag times (time between the user pressing the shutter button and the picture actually being taken) around 20 to 30 milliseconds. Digitals are amazingly worse than older mechanical cameras. I measured in my Hasselblad 500 ELM times between 32 and 40 milliseconds (with the reflex mirror locked up) depending on the lens (the V series cameras have the shutter in the lens); as Frans explains in his technique description, the delay on his Nikon D100 is a whopping 112 milliseconds. These times may not sound like much if you’re trying to capture your cat jumping out of a chair, but if the field of view is not much bigger than a bee, in that time the bee could almost completely leave the frame. Not only that, but it could stray from the focal field and appear blurred. At such high magnifications, it is difficult to get much depth of field (and sometimes short depth of field is desired). As you can guess auto-focus is out of the question, so the camera must be triggered right when the insect flies through the field of view at a depth in which it would be in focus. This is exactly what Frans has done, and his setup is carefully explained in his technique pages.

Aside from the shortest possible shutter lag, an extremely short exposure is also necessary to prevent the wings from appearing blurred. Normally in high-speed photography the exposure time is controlled by the duration of the illumination source; rarely is it controlled by the shutter. In other words, the room is left dark, the shutter left open, and a strobe, laser, spark, whatever, provides a short burst of illumination to create a short exposure of the subject. A typical Nd:YAG laser used for PIV has pulses about 10 nanoseconds long; the fastest industrial strobes are about 1000 times longer (10 microseconds). The Sunpak 622 Super Pro, which is arguably the most powerful handle-mount flash (and very cheap to boot) claims a minimum pulse duration of 1/30,000 seconds, or 33 microseconds. (There is a service manual for the 622 here.)

Whatever the duration of the light source, it must deliver enough light during this time to properly expose the picture, which normally means the power must be quite high. I’ve not yet looked at it carefully enough but I doubt that a pulsed laser powerful enough to use for this is either affordable or portable. Lasers do have the advantage that they are extremely easy to manipulate, that is, the beam can be made any size (within reason) relatively easily. Strobes and flashes are much cheaper, and although their pulse minimum pulse widths are higher than a laser’s by quite a bit, the length is still adequate. The problem is that strobes’ pulse widths are proportional to the amount of light they emit, so they can only have these extremely short pulses at the lowest power setting. Paying thousands for an industrial/scientific strobe probably solve that to some extent, but that’s not something I consider feasible. Some photographic flashes, including the 622, have shorter pulse widths in automatic mode, where a switch sending the light bouncing off the subject shuts off the current to the flash bulb “mid-stride”, than in manual mode, where the pulse width is simply a consequence of lowering the power sent to the flash bulb. So the way to guarantee the absolute shortest pulse is to put the flash in automatic mode and reflect some of the light straight from the flash into the light sensor (thyristor). Presumably a continuous bright light would do the trick, too, but I haven’t measured how much you need and I don’t know if that is bad for the thyristor in the long run.

To give an idea of what these pulse widths mean, let’s consider the classic case of the bullet-through-the-apple pictures. According to this, the speed of bullets can range from 100 to 1000 meters per second or so. We’ll take 500 m/s, roughly equal to mach 1.5 at sea level. In 10 nanoseconds, the bullet will travel

500\times 10\times 10^{-9}

or some 5 microns. This astoundingly small distance is less than the pixel size of most digital sensors out there, which means that if a bullet were 9 mm in diameter, and you took a picture of it with a 1600×1200 Kodak CCD (KAI-202x, pixel size 7.4 microns) so that the bullet were half the image height in diameter (magnification of 0.5 or so) it would have moved 2.5 microns or about 1/3 of a pixel – probably a small enough distance to look plenty sharp on the final image. We can talk about the magnification in some more photographic terms if we recall the thin lens equation:

\frac{1}{f} = \frac{1}{s} + \frac{1}{i}

where f is the focal length, s is the distance from the lens to the object, and i is the length from the lens to the image. The magnification using the thin lens/pinhole optics assumption is simply the ratio of any two corresponding (between object and image space) distances, such as

M = \frac{i}{s}

so that the focal length is

f = \frac{M}{M+1}s

or in our case


which means the above case is saying that we are taking a picture of a bullet 900 mm away with a 300 mm lens – in other words, these are extremely high magnifications.

The case of our Sunpak 622 Super Pro in auto mode with light blasting into the thyristor, with a 33 microsecond pulse width (3,300 times longer than our hypothetical laser), if we assume that the light intensity is constant through this duration (which it is not) will allow the bullet to travel 3,300 times farther than the laser during the exposure, which comes out to be 16.5 mm – nearly two diameters. Pretty disappointing, huh? If we wanted a picture with the same sensor to be as sharp with this flash as it was with the laser, then the magnification would have to drop to 2.5 microns (distance that the image of the bullet travels on the sensor) divided by 16.5 mm (distance the bullet travels in space) which comes out to 0.00015. If we assume that a “normal” lens has focal length equal to the diagonal of the sensor, then for this sensor the focal length would have to be about 15 mm, and the bullet would have to be 99 meters away. The magnification alone tells us the 9 mm diameter translates to 1.3 microns (less than one pixel), so it really is impossible.

Insect wings move more slowly than bullets do (hopefully). Actually, not hopefully; I’ve never heard of an insect with supersonic flapping wings.

If the pictures are not to be taken in a dark room, then the effect that natural light has on the picture comes into play, so that even though we would rely on the short light pulse from a flash to get the shortest possible exposure, the total light from the flash which participates in the image must be substantially more than the natural light which enters the lens during the time the shutter is open. Photographically, the light from the flash has to be some number of stops brighter than the natural light at the given shutter time, and the number of stops depends on the film characteristics and conceivably human perception (we could have relaxed our bullet sharpness requirements from 1/3 pixel to, say, 10 pixels if we said the final image would be reduced by 50% and viewed on the internet, at which point it’s doubtful a 2-pixel motion blur would be that drammatic). More on this will come later.

In summary then, there are three factors to consider for the insect camera:

  1. The shutter lag must be minimized.
  2. The shutter time (exposure time of natural light through lens) should be minimized.
  3. Enough flashes must provide enough light together to properly expose the film at their fastest setting.

Considering these challenges, everything else will be a piece of cake!

About this entry