How does a camera in an automatic mode choose what exposure, aperture and ISO to use?

For example:

1. 
   

   Does the camera detect if the scene contains fast-moving objects and if so use a shorter shutter speed?

   
   


2. 
   

   Does it detect if the entire scene is beyond the hyperfocal distance, and there's plenty of light, and automatically use an aperture at which the lens is the sharpest? Put differently, does it use the sharpest aperture unless there's a reason not to?

   
   


3. 
   

   Alternatively, does the camera detect if I'm shooting a portrait and switch to the widest aperture to de-focus the background?

   
   

Is there any intelligence at all here? Can I rely on the camera to pick the right aperture the vast majority of time, like I can with metering, and with focus when there's enough light?

Obviously, when I want to use a particular setting for a certain creative effect, I specify that, and not leave it to the camera. For example, if I'm shooting a train under low-light, I might choose a several second exposure to make the train appear milky, or a 1/4 - 1/2 second exposure to create some blur while still retaining distinct objects to some extent, or an even shorter exposure to freeze the motion. Obviously, I can't expect the camera to guess what I have have in my mind. This question is not about these cases. It's about cases where there's an obvious default that's right the majority of time. Can I rely on the camera to pick these?

In case the answer depends on the camera, this is with reference to the Sony NEX-5R.

When I say "an automatic mode", I mean automatic, program, shutter-speed priority and aperture priority modes. In the latter two modes, the camera still has to choose between ISO and aperture / shutter speed.

Answer

Camera manufacturers use their own proprietary algorithms for exposure metering. I doubt anyone will be able to describe the exact exposure and scene selection algorithms used by sony nex cameras without risking being sued by Sony, but one can make educated guesses so I will make such an attempt.

Sony states that in Intelligent Auto mode it performs Scene Recognition:

Scene Recognition operates in [Intelligent Auto] mode. This function lets the camera automatically recognize the shooting conditions and shoot the image

Based on that statement, it is evident that some kind of image analysis is done in order to determine the scene before the photo is taken. The scene in this case will determine whether to use aperture priority for macro and portraits or whether to use shutter priority for scenes like sports.

Not knowing the exact algorithm being used, but coming from an AI background, I'd guess that the algorithm most likely analyses the scene and attempts to detect certain parameters to use for scene selection such as object detection, object separation, rate of camera movement, and so on. Those parameters, in turn, will most likely be used as inputs to scene selection of some kind of fuzzy logic algorithm. The algorithm is most likely learned from the analysis of thousands of shots to approximate the ideal scene selection. The learned algorithm (assuming they use one) would depend on the quality, quantity, and diversity of the photos used for learning the algorithm. Normally, many photos will be analyzed, whose input parameters and the resulting scene and settings have already been detected manually. For simplicity, let's take an example of learning from 2 shots:

1. Inputs: Detected face, separation from background, small to no camera movement. Output: portrait mode, aperture priority, small aperture


2. Inputs: Fast camera movement. Output: sports mode, shutter priority

Now, when the camera detects a face, it will most likely choose portrait mode, if it detects fast camera movement then it will choose the settings from #2. The more images the algorithm is learned from, the better quality of the inputs, the more accurate the result will be.

Normally, algorithms like these have been learned before the camera hits the stores. I'm not sure if there are cameras that use adaptive algorithms where they learn based on your own camera usage what kind of shots you prefer and then appropriately skew the scene selection towards your preferences. I doubt self-learning mechanisms are used nowadays, but it certainly be sweet.

Now, the accuracy of the scene selection will depend on many factors:

• The ability to detect the inputs needed (such as object detection, face detection, object separation, camera movement, histogram, etc)


• The quality of the learned algorithm


• The time the scene is analyzed


• And many others depending on the exact algorithm used

It is safe to say, that there is lots of room for error no matter how intelligent the scene selection is. Assuming the scene selection is detected properly, then there will be additional algorithm to determine the proper exposure. The scene selection will most likely set limit preferences such as aperture between X and Y for portraits, shutter between A and B for sport scenes, and so on. Then the exposure algorithm will take those preferences into account and then complete the exposure triangle based on the light meter.

Now, no matter how good and sophisticated are the algorithms, there is no guarantee that it will choose the exact settings that you would have preferred. Therefore, photographers usually shoot in manual modes to have complete control over the image that they envisioned.

I guess it all depends what kind of shots you want. If you want quick snapshots, then use auto and be subject to the AI algorithms. If you want creative control, then use more manual modes.

If you want automatic control, but want to help the algorithms, then I'd suggest using scene selection to guide the algorithms by using sports, macro, portrait, etc. If you have controls to limit ISO levels then use those as well, depending on how much noise you are willing to tolerate.