Autofocus and burst: the specs that catch a moving subject

Modern autofocus has two halves. The first is measurement: phase-detection pixels on the sensor compare light arriving from opposite sides of the lens and work out which way, and how far, the focus has to move. That part is old and works well. The second half is newer and is what has changed photography in the last decade: a trained model looking at the live feed and deciding what in the frame is a subject.

In practice it means the camera finds a face, then the eye within the face, and keeps a box on it while the person moves. Most bodies now carry several detection classes — people, animals, birds, vehicles — and you pick one in advance, because a camera told to look for birds is more decisive about birds than a camera looking for anything.

What it really changes is where your attention goes. Older cameras made you place a focus point on the subject and keep it there, which meant composing around the autofocus. Detection removes that job, and the difference shows up as keeper rate rather than as anything visible in a single frame.

Detection is a pattern matcher, so it fails where the pattern is weak or ambiguous. Every one of these is routine rather than exotic.

• The subject is small in the frame. Below a certain size there is no face to find, and the camera falls back to whatever is nearest or most contrasty.
• The subject is facing away, or wearing a helmet, mask or sunglasses. Eye detection needs an eye; head and body detection are a coarser fallback.
• Something is in the way. Fences, branches, spectators, the glass of a vivarium. A branch across the bird is the classic wildlife failure.
• The background looks like the subject. Dappled woodland, a crowd behind an athlete, a patterned wall behind a face.
• The light is poor or the contrast is very low. Detection degrades before measurement does, and both degrade before your eye does.
• Two candidates compete. Two faces at slightly different distances, and the camera keeps changing its mind at exactly the wrong moment.

The practical response is to keep a manual override you can reach without looking: a small focus area you can place yourself, on a button. Detection is a very good default and a poor absolute. Note too that a faster lens makes every one of these failures more punishing, because depth of field is thinner at a wide aperture — which is the one real exception to the advice in lenses before bodies .

Burst rate is frames per second. The bodies here run from 3 to 40 fps. Buffer depth is how many frames the camera can take before it has to stop and wait for the card, and it is the number that decides whether you got the sequence — because the two divide into each other.

Eighty frames is an illustration, not a specification: real buffer depth depends on the file format, the compression, the resolution and how fast the card can be written. The point stands whatever the number. Doubling the frame rate halves the time you can hold the shutter down, so a high rate with a shallow buffer gives you a very fast two seconds and then a camera that ignores you while it writes. In practice the fastest bodies ship the deepest buffers, so the penalty is milder than the table alone suggests — but it is the pair of numbers that decides how long you can shoot, never the rate on its own.

There is a cost on the other end too. Twenty frames per second for five seconds is a hundred files to look at, and a session of that is an evening of culling. The people who get the most out of high frame rates shoot in short bursts timed around the moment, rather than holding the button down and hoping.

A mechanical shutter is two curtains crossing the sensor. An electronic shutter has no moving parts: the sensor is read out line by line. Electronic is silent, produces no vibration and is how the highest frame rates in this catalogue are reached. It also has one significant problem.

Because the lines are read in sequence rather than all at once, anything that moves during the readout is recorded at different moments in different parts of the frame. The result is rolling shutter: vertical lines lean when you pan, propellers and spokes bend, and a golf swing arrives visibly distorted. How bad it is depends entirely on readout speed, which is a sensor property the specification sheet rarely states plainly — stacked sensors read fast enough that it stops mattering, and cheaper ones do not.

• Artificial light flickers. Under LED or fluorescent lighting a slow electronic readout produces banding across the frame, and a mechanical shutter or an anti-flicker mode is the fix.
• Flash sync usually needs the mechanical shutter, and works only up to a stated speed, commonly in the region of 1/200 s to 1/250 s. Bodies with the fastest readout can now sync electronically too, so check the model rather than assuming.
• Mechanical shutters wear out. They are rated in actuations and replaceable, which is a running cost worth knowing if you shoot high volumes.
• Silence is a real feature, not a gimmick. Ceremonies, theatre, wildlife at close range and anywhere a shutter click changes what happens in front of you.

Following a moving subject is a feedback loop: you see where it went, you move the camera, you see the result. Blackout breaks the loop. On a slower body each frame of a burst interrupts the live feed, so the viewfinder shows a slideshow of the pictures you have already taken instead of where the subject is now. At ten frames a second that is tolerable for something moving predictably and hopeless for a bird changing direction.

Blackout-free shooting means the sensor is fast enough to keep feeding the viewfinder while it records, so the display never stops being live. It comes from the same readout speed that controls rolling shutter, which is why the two tend to be solved together and why they arrive at the same point in the price list.

Two related numbers are worth reading. Viewfinder refresh rate — whether the display updates at 60 or 120 frames a second — decides how smooth a fast pan looks. And display lag decides how far behind reality the image is, which your hands will quietly compensate for if it is small and cannot if it is not.

Autofocus and burst are the specifications marketing leads with, because they are the ones that produce a number that got bigger. Four groups genuinely need them.

• Sport, where the decisive moment is somebody else's timing and the subject is moving across the frame at speed.
• Wildlife, especially birds, where the subject is small, erratic, partly hidden and will not be repeating the performance.
• Children and dogs, which is where most people actually discover their camera cannot keep up. This is the largest group by a distance.
• Events and weddings, where you cannot ask for a second take and the light is usually poor.

Everybody else is buying insurance. Landscape, architecture, still life, food, studio portraits and most street photography are single-frame disciplines where the subject waits, and where a body at 5 fps with competent eye detection has never been the limiting factor. If that is your list, the money belongs in glass — the case is set out in what to spend , and how to choose a camera works out which camp your photographs are in.

One qualifier before you rule it out. Subject-detection autofocus has become genuinely useful for static work too: eye detection at f/1.8 on a portrait is more accurate than most people are with a focus point, and it removes an entire category of near-miss. That capability is present well below the top of the market, and it is not the same purchase as 40 fps. Compare bodies on both rather than taking the headline rate as a proxy for the autofocus.