Professionals thrive on feedback. An amateur can shoot at a deer and know he hit the right spot if that deer falls down. If the scope is zeroed right and the amount of compensation was correct, if the rifle has adequate accuracy for the range, the deer will fall. If not, the hunter can try again or go to lunch to a restaurant.
A professional dealing with a variety of targets, from deer to enemy soldiers to dinosaurs escaped from secret labs would more likely use a machine gun with tracers to adjust aim by observation of previous shots. That way the exact sight adjustment can be imperfect but the results would still be good enough and more reliably obtained.
By that logic, studio professionals have used Polaroids for decades. If, despite their best efforts, the lighting or other factors were not perfect, the pros could refine their aim with successive “instant” captures before proceeding onto regular film. Think “ranging machine gun coaxial to tank cannon” approach.
With cameras, we see a curious inversion of this principle. Currently, professional cameras use phase detection auto focus and amateur cameras use contrast detection auto focus. Phase detection is faster and a bit more reliable in low light, while contrast detection is more precise. That’s because DSLR focusing is conducted by a separate system which is calibrated at the time of construction but generally not updated once the camera is in use. The imperfections and eventual changes in calibration wouldn’t even be noticed until the photos are reviewed under higher magnification, at least with conventional lenses. With pro grade large aperture lenses for low light and nature photography, the margin for error is tiny due to the minimal depth of field. This is where contrast detection from the sensor could theoretically shine — that system provided incremental feedback directly from the final image.
At this time, contrast detection isn’t perfect or even good enough under certain circumstances (such as low light or fast subject movement), but the on-going improvements make me wonder if video viewfinders, the traditionally amateur feature, wouldn’t become common on pro cameras before long. After all, they are standard on the extremely expensive video cameras and that doesn’t surprise anybody.
If phase detection is superior in principle, I wonder if AF calibration can be incorporated into the start-up routine for every serious camera. Somehow, the concept of serious equipment with one of its most important functions on an open loop without internal feedback from the result seems like a stop-gap solution to me. Am I missing some technical reason why phase detection AF is here to stay?
I don’t know about having it self-calibrate on every start-up (time, potential accuracy issues if the camera is bouncing around while it’s doing the procedure, etc.). But a method by which you can recalibrate the camera on-demand would be a good idea.
AF calibration is already available for cameras that incorporate SR in-body. And most DSLRs that offer some version of Live View have at least some rudimentary CD focusing capability.
The thing that will likely drive contrast detection forward is the development of more serious MILC cameras and more demand for video camera capabilities in higher end cameras, since in both cases you have to rely on the CD system with no mechanism to divert light into a separate path. The systems in the latest m43 and NEX systems get in the ballpark of DSLR speed. The problem is, with the exception of Nikon’s 1 system, none of the current MILC focusing systems can track moving subjects all that well.
The main advantage of phase detection is that one measurement gives you more information, enough to determine which way to turn the focusing ring and by how much. That’s why it’s open loop – normally there is no need for another correction. You can easily close the loop (by switching to AI Servo), or keep it open.
Contrast detection gives you a measure of how blurry the image is, but each such measurement by itself is not enough to focus. You need to rack the focusing ring back and forth, taking successive measurements and looking for the point where the blurriness is at the minimum.
For the same reason split prism focusing screen is faster than just a ground glass.
Film is NOT dead.
I noted some years back that people who had it should use contrast-detect focusing (or manual focusing) for lens testing, if their camera had live view, precisely because it’s more accurate.
Some P&S and EVIL cameras are starting to integrate a few phase-detector elements on the sensor, so they can do faster focusing without having the complexity of a mirror system. And they’re advancing the art in contrast detection AF as well. Soon they’ll be good enough for most people I think.
What I’m worried about is that the big middle ground of semi-pro cameras that I’ve lived most of my life in is going to go away, leaving only consumer cameras and super-expensive (since sold in small quantities) professional cameras.
I’m a total photo neophyte, but I dig tech. I assume both of these systems make calibration decisions based upon some complex algorithms and from input from the sensors. Why not combine these systems? Wouldn’t the appropriate combination give you the “best of both worlds”? Or would it just be a mediocre shot based upon trying to satisfy both types?
Then there is a completely different scheme, which I recently saw described in a wall st. journal article. Interesting. If it really works, it’s the largest innovation in cameras since Daguerre.