[maxxxx]

This is a follow up to my post about how the camera chooses focus points (nobody really knows for sure).

The next question is regarding tracking. A lot of reviews write about how good or not good the tracking is but I can't figure out what that really means. Does it mean to adjust the same focus point when the object comes close or moves away? Or is it about moving the focus from one to another focus point as the object is moving across the frame? In Liveview with contract detection it's quite easy to see how the camera latches on to an object and tries to keep it in focus while it moves. But with the view finder and phase detect I can't figure out how it's supposed to behave and how it really behaves. Sometimes it seems very random.

Does anybody know how the tracking works or can point to an article or book? what's the difference between Nikon (who are supposed to be better) and Pentax tracking?

Cheers!

[MetteHHH]

There is a youtube video here, explaining how it works, and what is new to the K-3 system:

You Tube

You Tube

(If you are like me, you would prefer an article or book to a video - but all I saw were videos).

There is another video somewhere on youtube showing a bird flying behind trees, which was very encouraging for me, since I shoot birds and am currently saving for a K-3!

[Adam]

This is a follow up to my post about how the camera chooses focus points (nobody really knows for sure).

The next question is regarding tracking. A lot of reviews write about how good or not good the tracking is but I can't figure out what that really means. Does it mean to adjust the same focus point when the object comes close or moves away? Or is it about moving the focus from one to another focus point as the object is moving across the frame? In Liveview with contract detection it's quite easy to see how the camera latches on to an object and tries to keep it in focus while it moves. But with the view finder and phase detect I can't figure out how it's supposed to behave and how it really behaves. Sometimes it seems very random.

Does anybody know how the tracking works or can point to an article or book? what's the difference between Nikon (who are supposed to be better) and Pentax tracking?

Cheers!

Tracking just refers to the camera's ability to accurately predict the motion of the subject within the AF area and stay focused onto the subject.

This may be an oversimplification, but a key thing that makes for successful tracking is a. the rate at which camera camera samples the AF points for movement & adjusts the focus and b. the number of AF points. If you have a large, dense AF point grid that spans most of the viewfinder, it makes it easier for the photographer to keep the camera pointed at the subject and easier for the camera to predict where the subject is due to a larger number of data points.

It's hard to explain but if you pick up one of the more advanced Canon or Nikon DSLRs you'll see how they do it differently from Pentax. And on top of that, the speed of the lens AF motor is just as important as the camera's algorithm so the performance really depends on the lens.

[stevebrot]

But with the view finder and phase detect I can't figure out how it's supposed to behave and how it really behaves.

It does not behave so much as it misbehaves.

I think that it is a reasonable generalization that if a person is dissatisfied with AF of stationary objects, they will be totally disgusted with predictive AF. There is, of course, the odd Nikon owner out there and they seem to be almost universally thrilled with the AF on their cameras. Go figure.


Steve

(...must get tongue unstuck from cheek...)

[maxxxx]

There is a youtube video here, explaining how it works, and what is new to the K-3 system:
PENTAX K-3 ?Auto tracking? - YouTube

(If you are like me, you would prefer an article or book to a video - but all I saw were videos).

There is another video somewhere on youtube showing a bird flying behind trees, which was very encouraging for me, since I shoot birds and am currently saving for a K-3!

This is quite interesting, although very short. It seems it demonstrates that having the focus points spread out over a larger area makes a lot of sense for tracking.

Does anybody know if there is a way to find out which focus point was used for a certain shot? I guess looking at the focus points that were used while shooting a sequence would teach a lot.

---------- Post added 02-24-14 at 06:52 PM ----------

This is quite interesting, although very short. It seems it demonstrates that having the focus points spread out over a larger area makes a lot of sense for tracking.

Does anybody know if there is a way to find out which focus point was used for a certain shot? I guess looking at the focus points that were used while shooting a sequence would teach a lot.

Just checked with PhotoME and found that the pictures have quite a few properties regarding auto focus. For example "AF Points In Focus", "AF Predictor", "AF Integration Time". Maybe if I find time this weekend I'll find a way to visualize this information.

[pacerr]

This is quite interesting, although very short. It seems it demonstrates that having the focus points spread out over a larger area makes a lot of sense for tracking.

I have a hard time using that video as an explanation as to WHY multi-point AF is better than single point. The video maintains a fixed view of the subject area which makes tracking the nearest element of the subject as it crosses the frame seem logical. However, were I making that shot I'd likely be panning to keep the subject at about the same point of composition within the frame such that a single AF point might be the logical point of focus for the subject.

It only makes sense to me if one assumes that the nearest point in focus within the AF array is always the desired point of focus in the scene. And that would only work well if the camera was capable of processing speed and lens reaction time suitable for accurately predicting a continuous focus solution on any moving target.

[maxxxx]

I have a hard time using that video as an explanation as to WHY multi-point AF is better than single point. The video maintains a fixed view of the subject area which makes tracking the nearest element of the subject as it crosses the frame seem logical. However, were I making that shot I'd likely be panning to keep the subject at about the same point of composition within the frame such that a single AF point might be the logical point of focus for the subject.

It only makes sense to me if one assumes that the nearest point in focus within the AF array is always the desired point of focus in the scene. And that would only work well if the camera was capable of processing speed and lens reaction time suitable for accurately predicting a continuous focus solution on any moving target.

If you CAN pan in a way that the subject is always at the same point in the frame then single point is probably better. But with fast moving subjects this is often not possible so having the capability that the focus moves from one point to the next is helpful. I am just not sure the camera really works that way..,

[Cold Iron]

There is a youtube video here, explaining how it works, and what is new to the K-3 system:
PENTAX K-3 ?Auto tracking? - YouTube

(If you are like me, you would prefer an article or book to a video - but all I saw were videos).

There is another video somewhere on youtube showing a bird flying behind trees, which was very encouraging for me, since I shoot birds and am currently saving for a K-3!

I have had my K-3 for a month now my first DSLR and know just about enough to be dangerous at this stage.... I too shoot Birds and am using single point focus and tracking although several threads on here have explained that multi-point is better if say a wing drops you could loose focus. I set mine up Single Point tracking more by mistake than anything under U1 but have a hard time convincing myself to change it. Much of my shooting is in heavily wooded areas with small windows to shoot through and have to be quick to lock on especially with the branches in the way. I have been practicing with geese as they head back out of the fields in the evening with low light.



Tracking into the trees has not been an issue yet for me. Here the sun was still hitting the geese turning the white orange on them as they flew above me through very dense trees and it was pretty dark where I was.


And here a whole flock flew behind a LOT of trees, I was shocked that it kept focus the whole time


After shooting in the trees and narrow tight windows a bird in the open with clear bluebird skies is almost a give me.


I'm still working on it but I think you will really enjoy the K-3. To me it has been well worth it.

[starbase218]

I think that "tracking" most often means that the camera can track a subject across the frame. This is accomplished by feeding information from the light sensor to some kind of algorithm which then decides if an adjacent focus point should be activated in order to keep the subject in focus. I think this is probably one of the reasons the new light sensor on the K-3 is an 86K pixel sensor.

Adjusting the focus when the subject moves closer or farther away is something Pentax (and other manufacturers) have been doing for ages. In case of Pentax, an algorithm continuously checks whether the focus under the focus point is still locked. If not, it will refocus the lens. This process is repeated as long as the shutter is half-pressed. I see no reason why this could not be combined with the tracking as described above. So, the light sensor is checked to select the correct AF point(s), and focus is continuously checked on those selected points.

For both systems, one could imagine some kind of "predictive" algorithm that takes the last few adjustments into account.

[stevebrot]

I think this is probably one of the reasons the new light sensor on the K-3 is an 86K pixel sensor.

I think that refers to the viewfinder light metering system and is not related to the AF system. The PDAF system (under discussion on this thread) is still at the bottom of the mirror box and uses the 27 focus points mentioned earlier.


Steve

[derekkite]

I have a hard time using that video as an explanation as to WHY multi-point AF is better than single point. The video maintains a fixed view of the subject area which makes tracking the nearest element of the subject as it crosses the frame seem logical. However, were I making that shot I'd likely be panning to keep the subject at about the same point of composition within the frame such that a single AF point might be the logical point of focus for the subject.

It only makes sense to me if one assumes that the nearest point in focus within the AF array is always the desired point of focus in the scene. And that would only work well if the camera was capable of processing speed and lens reaction time suitable for accurately predicting a continuous focus solution on any moving target.

There are two functions from what I can see. The extended area af will start with the selected point and keep focus on the subject. I suspect that it is by a few different means, using a contrast shape or color, and some rate algorithm to determine movement closer or farther. It does that reasonably well, not prescient but consistent. It has trouble if there isn't anything of notable contrast to work with due to either low light or simple low contrast. A simple test is to focus the initial point on a subject about the size of a focus point, then pan slightly within the extended area points and it will follow, you can see the other points twinkle as they are used. The K-5 would select the highest contrast within the points.

This would be the foundation of a tracking system, and if the focus points are accurate and there is adequate contrast, it works well. The K-5II had this functionality from what I understand.

The K-3 has another function which adds a bias to keeping focus on the initial subject focus distance. In the simple test above, if you pan the object past the extended area points it will refocus on something closer or further away. The hold setting of low, medium or high will bias a tendency to not refocus. It seems that the longer you have a subject in focus the more the bias towards not refocusing. If you focus on something, holding the AF or half shutter, then immediately pan to something else it won't hold. This is common usage to find the point on which you want to focus. But if you focus, maintaining AF or half shutter button, hold it there for a bit then move the refocus will be delayed.

The video example indeed could be easily replicated by panning technique, but for example tracking a bird in flight. It is very hard to keep them accurately lined up for single point focus, so extended area works nicely. If you lose it as it changes direction or something comes between you and the subject, the hold allows you to keep within the focus distance without the lens hunting back and forth and you losing the shot altogether.

It isn't perfect, I think the next step is to integrate recognizable shapes such as eyes, face or head profiles. The Pentax isn't smart enough to pick out the subject you want to follow (at least in PDAF). But it is predictable. In a given scene, or a similar scene it will do the same thing, which allows me the photographer to get to know the limitations and strengths and work with them for good results. That was not the case with the K-5, it was unpredictable. I experienced that the other day, with my K-3 in for repairs. Perfect light, late afternoon through a high thin cloud generating warm high contrast light, moments we wait for. Swans not too far away doing what swans do, and there would be 3 shots in perfect focus, two not, 4 good, a couple not. The K-3 in similar situations produced one shot after another focused just right.

A technique that K-3 users will learn is to touch and release the focus button to reset the hold.

The extended area focus works for focus and re-frame without the angle changes losing focus.

[starbase218]

I think that refers to the viewfinder light metering system and is not related to the AF system. The PDAF system (under discussion on this thread) is still at the bottom of the mirror box and uses the 27 focus points mentioned earlier.


Steve

I know and it is related to the AF system. What I'm trying to explain is that, with 86K RGB pixels, the light metering sensor can be used to provide information on where the subject is going, so that the camera can select an adjacent focus point if the subject is moving there.

Nikon has done this for ages with 2016 pixel RGB sensors. See http://imaging.nikon.com/lineup/dslr/d7100/features01.htm.

[lister6520]

I know and it is related to the AF system. What I'm trying to explain is that, with 86K pixels, the light metering sensor can be used to provide information on where the subject is going, so that the camera can select an adjacent focus point if the subject is moving there.

Nikon has done this for ages. See Nikon | Imaging Products | Capturing power - Nikon D7100.

Not only can be used but is actually used for that purpose. It was described in the pre release videos of the K-3. The 86K exposure sensor is effectively a digital video camera in its own right, just a very low resolution one. If the engineers at Pentax wanted to they could use it for all sorts of things that are normally only available in live view, such as face detection, scene analysis for selection of auto mode and other such stuff, though most of that would be stuff you would want on a P&S or entry level DSLR. Focus tracking is however the exception where it can contribute enormously to having an 'intelligent' focus tracking system and according to Pentax's description it does.

In use I have found it to work very well as opoosed to what some people seem to suggest or even declare.

The earlier 'expanded area' AF introduced with the K-30 had to make do with the information from just the nine focus points. The algorithm had to try to make sense of what's happening based on very limited information. It still works nicely in some situations but in others it cannot cope at all. The K-3's system on the other hand seems to almost always work fine.

[stevebrot]

Not only can be used but is actually used for that purpose. It was described in the pre release videos of the K-3.

Ok. I did not see the pre-release videos and have not read anything about this feature in post-release reviews, but I will take your word for it. If this is the case, what happens to focus tracking when using a non-Pentax aftermarket screen having focus aids?


Steve

[starbase218]

Ok. I did not see the pre-release videos and have not read anything about this feature in post-release reviews, but I will take your word for it. If this is the case, what happens to focus tracking when using a non-Pentax aftermarket screen having focus aids?

I guess it could get screwed up. The light meter sensor looks at the image on the focusing screen. Even spot metering gets it wrong on my K-5 when using such a focusing screen. But why would you use a focusing screen with manual focusing aids when using tracking autofocus (except that you don't want to continuously switch focusing screens)? These are really two very different approaches to focusing in general. You can still disable tracking and do it the old-fashioned way. But I guess that that improvement to the K-3 won't be of interest to you, then.