ClarkVision.com
| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
Autofocus Speed of DSLRs and Telephoto Lenses
by Roger N. Clark.
| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
by Roger N. Clark.
Autofocus speed and accuracy is crtitical for fast action photography, including wildlife, birds in flight, sports, and even children and pets at play.
The Action Photography Series:
Contents
Introduction
Methods
Camera Configuration
Focus: Near to Far versus Far to Near
Analysis
Results, Method 1
Results, Method 2
Discussion and Conclusions
Appendix 1
References and Further Reading
Introduction
With this article, I'll present measurements of the autofocus speed of some camera and lens combinations. There is very little information on the speed of autofocus of cameras, whether from manufacturers, or from review sites. There is, of course, a lot of variability in conditions in which one photographs where autofocus speed is critical in obtaining the image. But the basic speed at which a camera can move a lens sets some upper limits. I have done many tests, each with the same conclusion. I also photograph varying subjects in action, ranging from African wildlife, North American wildlife, birds in flight, and pets at play.
The toughest subject for autofocus to lock on and track is a dark subject on a varied and brighter background. Figure 1 shows a very fast flying boat-tailed grackle that was a particularly difficult subject to track. Small birds generally fly very fast and one needs to be at relatively close range if they are to fill much of the frame. Small birds are not only difficult for the autofocus system to track, but they are difficult for photographers to track with the small field of view with a long telephoto lens. The image in Figure 1 was made with a 500 mm f/4 lens mounted on a Wimberly gimbal head. Figure 2 shows a very fast flying lilac breasted roller made hand held with a 300 mm f/2.8 lens.
Wildlife and domestic pets can also be a challenge for autofocus to track. Figure 3 shows a black dog with a brighter background running straight at the camera at about 8 meters per second (distance = 12 meters) with a complex background. Note the eyes are in focus: I used one AF point off-center and on the eyes to track the subject.
I have over 10 years of experience of photographing action with super telephoto lenses and DSLRs, and also with older film SLRs and other lenses, in a variety of environments, from sports action, pets in action, wildlife and birds ranging from Africa to Alaska to Australia. The results below agree with my in-the-field experience. I currently own and use (sometimes side by side) Canon 500 mm f/4 and 300 mm f/2.8 super telephoto lenses (both version 1). But I am using the 300 f/2.8 more because of its speed and versatility in the field. For the last couple of years counting only telephoto lens use, I use the 500 mm about 1/3 of the time, and the 300 2/3 (statistics on my last 37,000 images).
I am finding the 300 f/2.8 lens has faster at autofocus than the 500, even with TCs. This has caused some controversy among some photographers. Canon has used the 300 f/2.8 in their published specs on autofocus speed. Some of their old literature called the lens the fastest autofocus (AF) lens, though recently their literature simply says fast AF so we are given little information on how fast lenses and TCs really are. Canon used the 300 f/2.8 in charts of AF speed. For example see Figure 5 at: http://www.canon.com/camera-museum/tech/report/200304/200304.html. In 2004, Canon published "Getting the Most from your EOS-1 Class Digital SLR" and Figure 2 shows "Predictive AI Servo AF Performance" using the 300 f/2.8 lens on a Canon EOS-1 film camera and a 1D Mark II DSLR, with the DLSR slightly beating the film camera. I can not find this document online anymore (I have the chart), but perhaps one might find it through this link: http://www.robgalbraith.com/bins/content_page.asp?cid=7-6460-7335. Historical documentation I have from Canon says the Canon 1D, D30 and D60 had the same predictive AI servo AF performance, and the 10D, 20D, 1D Mark II, 1D Mark III, 1Ds Mark II, and 1Ds Mark III had the same same performance but faster than the older group. In Canon's Canon 1D Mark IV white paper, EOS1D_MarkIV-WP1.pdf, available from http://www.learn.usa.canon.com/resources/products/eos_1d_markiv/eos_1dmarkiv_white_papers.shtml, page 15, Canon says the tracking speed of the 1DIV is the same as the 1DIII, but software upgrades mean better maintaining of focus lock in complex environments. I have extracted some data from that literature and show derived values in Table 1. Note that the data in Table 1 is tracking performance in an ideal situation, not necessarily the locking on performance, nor tracking ability in a complex environment with distracting elements, often found in wildlife action photography. If you know of plots with newer cameras, please let me know.
From Canon's published figures, we can derive several things. For example, the Canon 10D could track an object at 72 km/hour at a distance of 10 meters. If the distance is doubled, the object maximum tracking velocity would increase 4 times. (from the first link above, we see the performance trend on the log-log plot is a straight line with a slope of one half.) A velocity of 72 km/hour equals 20 meters per second. But the real quantity from a camera perspective is the rate change of the focus distance at the sensor (or the autofocus sensors). A simple computation of focal distances (see Appendix 1 for details of this calculation) leads to the following derived values using the Canon published figures (Table 1).
Table 1
Canon Predictive AF Focus-Tracking at 10 meters Focus-Tracking
Camera with Canon 300 f/2.8 L IS version 1 Velocity at
----- Canon Published Values ---- focal plane
Model km/hour meters/second mm/second
---------------------------------------------------------------------------
1D 40 11 10.6
D30 40 11 10.6
D60 40 11 10.6
10D 72 20 19
20D 72 20 19
1D Mark II 72 20 19
1D Mark III 72 20 19
1Ds Mark II 72 20 19
1Ds Mark III 72 20 19
1D Mark IV 72 20 19
---------------------------------------------------------------------------
So just how does the AF performance compare with other lenses and newer cameras? Here is a start to my testing.
Methods
Test Method 1. Target Acquisition Speed This method determins the speed of acquiring a subject. I use a right angle finder on the DSLR, and attach a web cam to the finder eyepiece. I record video as follows, alternating focusing on near and far targets. Except for the move, I do not touch the camera. Focusing is initiated by cable release so as not to shake the camera.
Analysis. Examine the video frame by frame. Determine the first frame where the focus starts to move, and then find the frame when the green focus confirmation light lights and/or the mirror starts to lift. The time between these two events is the time I report to focus. At 30 frames per second, relative timing is accurate to +/- 0.033 second, or for the start + stop: square root 2 times 0.033 or +/- 0.05 second for the interval.
Test Method 2. Alternating Target Acquisition Speed To simulate the "heat of battle" alternately rapidly move between two targets, one near, one far, framing as rapidly as possible. There is a human element in this test involving the movement. I placed the near and far targets one above the other, separated by about 2 degrees and did the tests using a Wimberly gimbal head where I only had to move up and down. To increase speed, I places a padded stop on the far target to be able to stop more accurately (this is after testing with no stop). Obtain an image or short burst at each position and alternate between positions as rapidly as possible. With some practice, I have been able to get my time to move between the targets down to about 1/4 second.
A variation on the methods is to use targets that sway in the wind. That simulates an erratic movement, like a darting animal or bird in flight. I have conducted the methods using static targets and wind moving targets. The wind-moving targets of course have more scatter in the timing results but point in the same direction of which are the fastest combinations.
How are these tests relevant? Both tests simulate acquisition of a subject. The camera must drive the lens as fast as possible to the desired focus point. Tracking, once acquired, is slower unless the subject is moving at the fastest the lens+camera is capable of moving, thus acquisition speed represents the fastest possible the system is capable.
Of course, many factors can influence results. For example, light levels. As light levels drop, AF speed and accuracy of locking on drops. That is evident in my data, but the relative positions of lens A +camera B versus Lens C on Camera B appear to remain the same, though with more overlap in the performance as performance degrades.
The distances used correspond to my real-world imaging of action. Table 2 shows image range data (in meters) and the number of frames imaged using super telephoto lenses in AI servo mode for five different sessions on safari. A subject-rich environment like that found in Tanzania is a constant choice of changing what one is imaging, often needing a fast response to quickly acquire a subject and track it. Thus, the tests simulate that acquisition of subjects. In my experience, if I can get to camera to lock onto the subject, and if I can keep the autofocus sensor on the subject, the camera is able to track it. So again, this target acquisition and lock test is one of the hardest parts of the imaging problem (in my experience).
Table 2
-------------------------------------------------------------------------
Serengeti Lake Manyara Tarangire N. Park Tarangire N. Park
-------------------------------------------------------------------------
afternoon morning morning afternoon
# meters # meters # meters # meters
2 4.83 2 3.03
41 8.465 2 3.29 1 2.81 4 6.735
5 9.735 6 3.61 2 6.12 2 7.495
22 11.48 3 3.8 4 6.735 2 8.465
79 14.035 188 4.01 10 7.495 5 9.735
142 18.145 16 4.245 21 8.465 24 11.48
52 25.92 65 4.515 16 9.735 4 14.035
53 31.62 49 4.83 25 11.48 20 18.145
414 47.18 29 5.19 24 14.035 60 25.92
10 6.12 59 18.145 81 31.62
13 6.735 85 25.92 451 47.18
10 7.495 185 31.62
14 8.465 333 47.18 Morning
5 9.735 # meters
42 11.48 2 5.19
12 14.035 3 5.61
68 18.145 2 6.12
155 25.92 6 9.735
87 31.62 3 11.48
29 47.18 6 14.035
29 18.145
56 25.92
102 31.62
268 47.18
-------------------------------------------------------------------------
Note that the ranges are quantized. This shows the quantization limits
of the lens.
Figure 4 illustrates the need for fast acquisition and tracking. The crowned hornbill was flying through the trees. Light levels were low and I was photographing another subject. I saw the bird, quickly turned to catch it, the AF locked, began tracking and I got this dark bird flying in the trees. Without the extremely fast AF of the 300 f/2.8, and a fast camera, this image could not be obtained. Note the sharp focus on the eye. Yes, the background is cluttered with all the branches; the point is AF performance.
Camera Configuration
In the tests below, I show two cameras and 3 lenses. All tests done in AI servo. The configurations are:
1D4:
December, 2014 update: initial testing of the new Canon EOS 7D Mark II DSLR Camera and Canon EF 300mm f/2.8L IS II USM Lens shows the combination provides similar autofocus performance as the Canon 1D Mark IV and 300 mm lens shown below. Quantitative numbers to be added.
Focus: Near to Far versus Far to Near
There is a marked difference in AF performance when the lens is moving from far to near versus near to far. Moving far to near is observed to be generally faster. Usually, we photograph subjects moving toward us, so if there was an asymmetry in AF speed, it is best for most cases that the speed is faster for subjects moving toward us. Moving from near to far usually happens when one is photographing a close subject and one switches to a far subject, such as an approaching animal or flying bird. Thus, in my opinion, the far to near numbers are more important. However, Figure 4 is a case where the initial acquisition was near to far, then the AF system had to switch to tracking the approaching bird. The test show that the 1DIV + 300 f/2.8 has the same fast tracking from near to far and far to near. More on this topic in the results and conclusions.
The right column of the results shows the calculated focus velocity for the far to near focus change and the measured time. Appendix 1 shows the lens calculation for computing the focus position. If the camera can track a subject at these rates, they could be compared to the velocities in Table 1. (I believe they can)
Analysis
All data were analyzed on linux using available linux tools. I wrote custom shell scripts to extract exif data from the images using exiftool for linux.
Results, Method 1: Target Acquisition Speed
close target = 5.0 meters
far target = 46 meters
Focus Velocity Subject Velocity
for an approaching at 10 meters distance
Camera lens TC f/ratio 46 to 5 5 to 46 subject mm/sec. km/hr m/s
------------------------------------------------------------------------------------------------
1DIV 500 f/4 - 4 0.47 0.61 110 142 39.5
1DIV 500 f/4 1.4 5.6 0.57 0.60 180 114 31.7
1DIV 500 f/4 2 8 1.10 0.65 200 58 16.2
1DIV 300 f/2.8 - 2.8 0.20 0.20 86 321 89.1
1DIV 300 f/2.8 1.4 4 0.40 0.43 92 171 47.6
1DIV 300 f/2.8 2 5.6 0.45 0.45 160 141 39.1
7D 300 f/2.8 - 2.8 0.38 45 169 46.8
7D 300 f/2.8 1.4 4 0.38 0.42 96 179 49.7
------------------------------------------------------------------------------------------------
Results, Method 2: Alternating Target Acquisition Speed
AF test fast moving between two targets.
Move from one target to the next, and get an image, using
AI servo, focus priority.
close target = 6.0 meters
far target = 20.0 meters
Higher light level:
------ Time in Seconds ------
mean st dev mean st dev Number of
Camera lens TC f/ratio 20m to 6m 6m to 20m observations Exposure
----------------------------------------------------------------------------------------------
1DIV 500 f4 - 4 0.65 +/-0.03 0.63 +/-0.04 62 1/1600 f/5.6 ISO 400
1DIV 500 f4 1.4 5.6 0.65 +/-0.03 1.14 +/-0.10 44 1/1600 f/5.6 ISO 400
1DIV 500 f4 2 8 1.08 +/-0.38 1.72 +/-0.53 56 1/800 f/8 ISO 400
1DIV 300 f2.8 - 2.8 0.47 +/-0.03 0.49 +/-0.06 57 1/800 f/5.6 ISO 400
1DIV 300 f2.8 1.4 4 0.59 +/-0.05 0.49 +/-0.05 72 1/800 f/5.6 ISO 400
1DIV 300 f2.8 2 5.6 0.62 +/-0.04 0.60 +/-0.05 78 1/800 f/5.6 ISO 400
----------------------------------------------------------------------------------------------
Lower light level:
------ Time in Seconds ------
mean st dev mean st dev Number of
Camera lens TC f/ratio 20m to 6m 6m to 20m observations Exposure
----------------------------------------------------------------------------------------------
1DIV 500 f4 - 4 0.81 +/-0.06 0.81 +/-0.07 52 1/800 f/5.6 ISO 800
1DIV 500 f4 1.4 5.6 0.83 +/-0.02 1.41 +/-0.11 25 1/800 f/5.6 ISO 800
1DIV 500 f4 2 8 2.18 +/-0.89 2.80 +/-0.61 24 1/320 f/8 ISO 800
1DIV 300 f2.8 - 2.8 0.55 +/-0.02 0.57 +/-0.08 57 1/800 f/5.6 ISO 800
1DIV 300 f2.8 1.4 4 0.63 +/-0.08 0.58 +/-0.06 59 1/800 f/5.6 ISO 800
1DIV 300 f2.8 2 5.6 0.73 +/-0.07 0.88 +/-0.33 70 1/800 f/5.6 ISO 800
1DIV 300 f4 - 4 0.88 +/-0.20 0.74 +/-0.06 23 1/800 f/5.6 ISO 800
----------------------------------------------------------------------------------------------
7D 500 f4 - 4 1.13 +/-0.10 1.43 +/-0.09 24 1/500 f/5.6 ISO 800
7D 500 f4 1.4 5.6 1.05 +/-0.10 1.97 +/-0.58 51 1/500 f/5.6 ISO 800
7D 300 f2.8 - 2.8 0.62 +/-0.05 0.63 +/-0.05 55 1/500 f/4 ISO 800
7D 300 f2.8 2 5.6 0.88 +/-0.11 1.11 +/-0.12 33 1/500 f/5.6 ISO 800
7D 300 f4 - 4 0.74 +/-0.09 0.93 +/-0.29 124 1/640 f/5.6 ISO 800
----------------------------------------------------------------------------------------------
Discussion and Conclusions
December, 2014 update: initial testing of the new Canon EOS 7D Mark II DSLR Camera and Canon EF 300mm f/2.8L IS II USM Lens shows the combination provides similar autofocus performance as the Canon 1D Mark IV and 300 mm lens shown above. Quantitative numbers to be added.
The webcam video tests show the maximum speeds of target acquisition with the lens +camera combinations. The 300 f/2.8 with no TCs is more than twice as fast as the bare 500 f/4 on the 1D Mark IV for far to near target acquisition (and tracking), and 3 times as fast from near to far. This translates to significant better ability to acquire a faster moving subject at close distances. The 300 on a 7D with no TCs is also faster than a 500 f/4 on a 1D Mark IV. The pixel pitches between the two cameras (5.7 micron on the 1DIV versus 4.3 on the 7D means that the 7D+300 f/2.8 is like a 400 mm lens on the 1DIV, so while the 500 +1DIV gets 26% more pixels (linear dimension) on a subject, the AF is faster with the 7D+300. Most DSLRs also have special f/2.8 (and faster) AF sensors which provide increased accuracy. It is the speed of the 300, the f/2.8 ratio, and the light weight that puts this lens in a class ahead of the larger telephotos.
The method 2 results are similar to the method 1 results. In both tests the 300 + 2x TC provides comparable or slightly faster AF than a bare 500 f/4 on the 1DIV and on a 7D. Note however, there is a degradation in the lower light levels. Examine the +/- standard deviation numbers. These numbers are the standard deviation of the population (not of the mean). A larger number means the lens had more issues with AF with some tries either starting the wrong direction, or an overshoot. For example, the 500 f/4 on the 1D4 from far to near has a standard deviation of 0.03 for the bare 500 but 0.38 second for the 500 +2x TC in good light. In lower light the 500+2x TC on the 1DIV increases to 0.89 second, indicating more misses and longer to achieve focus. This also occurs with the 300 f/2.8, but the effect for the light levels tested were less of an problem on both the 1DIV and 7D.
Note too when comparing AF times, that the light levels changed during the test, a consequence of outdoor testing. The 300 f/2.8 sequence on the 1DIV was at a lower light level that that for the 500 f/4. That biased the 300 f/2.8 results to longer times. If the 300 f/2.8 test with the 2x TC were done at the same light level as the 500 f/4 bare, the 300 would look a little better.
There is a significant drop in AF performance with the 500 f/4 + 2x TC. Only a single sensor is used and AF only works at f/8 on 1D series cameras. This explains why the 500 f/4 is a poor performer for birds in flight with a 2x TC.
The 300 f/2.8 version I with a 1.4x TC is a stellar performer, and the newer Canon EF 300mm f/2.8L IS Version II lens is even better, and the new lens is my choice over a 500 f/4 bare unless I really need the reach. The light weight of the 300 (version 1 or II) is easier to hand hold and faster to move hand held or on a gimbal head than the bigger lenses. Unless one needs the reach of the 500 (or larger lens) the Canon EF 300mm f/2.8L IS Version II is a top choice. When I head to Africa on safari, I have taken the 500 on some trips, and the 300 on others. I have been so impressed with the 300, that I'll only take the 300 f/2.8 on future safaris (of course this is subject to change, but that is my plan). When I had cameras with larger pixels (like the 5DII or 1DII), then more reach can be important. But with sub-6-micron pixels, like in the 1D Mark IV and Canon EOS 7D Mark II DSLR Camera, I feel that the longer reach is not needed for the subjects I image at the distances I can achieve. The lighter weight of the 300 (even the heavier version I) has allowed me to get great images of lilac breasted rollers in flight in the Serengeti, something I tried but could never achieve with the 500 f/4.
Image quality of the 300 with TCs is also outstanding. One must carefully calibrate the microadjustment for the temperatures one encounters in the field for the best results.
A smaller lens has smaller elements with less mass. Thus, it is not surprising that the Canon EF 300mm f/2.8L IS lens is so fast. Longer lenses require a longer movement of the focal plane (of course with more reach), but that faster movement and more mass to move at a higher rate mean more difficulty in making fast AF performance.
I'll add other cameras and lenses as time and opportunity permit.
Appendix 1
For a subject at a distance A and lens focal length FL, the distance behind the lens where the subject comes to focus is given by the relation:
1/FL = 1/A + 1/B
or
B = A*FL/(A-FL).
Example values of back focal distance, B:
object back focal distance (mm) for various lens focal lengths (FL)
distance ------------------------------------------------------------
(meters) 300 mm 420 mm 500 mm 600 mm 700 mm 1000 mm
FL FL FL FL FL FL
2 352.94 531.65 666.67 857.14 1076.92 2000.00
3 333.33 488.37 600.00 750.00 913.04 1500.00
4 324.32 469.27 571.43 705.88 848.48 1333.33
5 319.15 458.52 555.56 681.82 813.95 1250.00
6 315.79 451.61 545.45 666.67 792.45 1200.00
7 313.43 446.81 538.46 656.25 777.78 1166.67
8 311.69 443.27 533.33 648.65 767.12 1142.86
9 310.34 440.56 529.41 642.86 759.04 1125.00
10 309.28 438.41 526.32 638.30 752.69 1111.11
11 308.41 436.67 523.81 634.62 747.57 1100.00
12 307.69 435.23 521.74 631.58 743.36 1090.91
13 307.09 434.02 520.00 629.03 739.84 1083.33
14 306.57 432.99 518.52 626.87 736.84 1076.92
15 306.12 432.10 517.24 625.00 734.27 1071.43
16 305.73 431.32 516.13 623.38 732.03 1066.67
17 305.39 430.64 515.15 621.95 730.06 1062.50
18 305.08 430.03 514.29 620.69 728.32 1058.82
19 304.81 429.49 513.51 619.57 726.78 1055.56
20 304.57 429.01 512.82 618.56 725.39 1052.63
21 304.35 428.57 512.20 617.65 724.14 1050.00
22 304.15 428.17 511.63 616.82 723.00 1047.62
23 303.96 427.81 511.11 616.07 721.97 1045.45
24 303.80 427.48 510.64 615.38 721.03 1043.48
25 303.64 427.18 510.20 614.75 720.16 1041.67
30 303.03 425.96 508.47 612.24 716.72 1034.48
35 302.59 425.10 507.25 610.47 714.29 1029.41
40 302.27 424.46 506.33 609.14 712.47 1025.64
45 302.01 423.96 505.62 608.11 711.06 1022.73
50 301.81 423.56 505.05 607.29 709.94 1020.41
60 301.51 422.96 504.20 606.06 708.26 1016.95
70 301.29 422.54 503.60 605.19 707.07 1014.49
80 301.13 422.22 503.14 604.53 706.18 1012.66
90 301.00 421.97 502.79 604.03 705.49 1011.24
100 300.90 421.77 502.51 603.62 704.93 1010.10
---------------------------------------------------------------------
References and Further Reading
More birds in flight:
http://www.clarkvision.com/galleries/gallery.bird_in_flight/
Wildlife Action Photography: Autofocus Tracking with Digital Cameras http://www.clarkvision.com/articles/tracking.action.with.autofocus
The Action Photography Series:
| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
http://www.clarkvision.com/articles/autofocus.speed
First Published January 22, 2013
Last updated October 11, 2015