Predictive Autofocus: Does It Work?
How to focus on fast-moving subjects
Autofocus is a great tool. It saves critical time required to focus on a subject, and in many circumstances it enables you to get pictures that you probably would have missed had you been focusing the camera manually. For example, I shot wildlife with a medium format (6x7cm) film camera for 25 years, and I can’t tell you how many pictures I missed because I couldn’t focus fast enough to catch fleeting moments in nature.
When subjects are stationary, autofocus is convenient but not crucial. You have time to focus on a grazing rhino in the Serengeti or a horse on a farm. However, what if you are confronted by a herd of horses galloping right toward the camera? Does normal autofocus work?
No, it doesn’t. By the time the camera focuses on the horses and you press the shutter, a fraction of a second has elapsed, and when the picture is taken, the horses will be several feet closer to you. If you are using a long lens and filling the frame with the charging horses, they will be almost sharp but not quite. Obviously, this is not acceptable.
Autofocus tracking (also referred to as predictive autofocus and shooting in AI servo) is a feature built into many digital cameras, and it is supposed to determine how fast the subject is moving toward or away from you and thereby hold it in focus as it moves. The theory is that at any point in time you can shoot and get a sharp picture. In my experience, this doesn’t happen. Predictive autofocus just isn’t that good if the fast-moving subject is coming toward or away from the camera.
The only time this feature works well is if the moving subject is going across your field of view from one side to the other. For example, a flying bird that flies left to right can be tracked successfully with AI servo. However, if the bird turns 90 degrees and flies toward the camera position, there is no way you can track it with confidence using any autofocus function.
This photo of the galloping horses in sunset light is an all-time favorite image of mine. It took two trips to France to get this. The first trip was made at the beginning of July to coincide with the lavender fields in Provence, but there wasn’t enough water in the marshes. In addition, my longest lens was a 200mm because I assumed the horses would be somewhat staggered as they ran toward me, and I was hoping that the fairly good depth of field characteristic of 200mm lenses would keep all the horses sharp even if they were staggered a bit. This was a mistake. The reason is I had to wait for the horses to fill the frame as they galloped toward me, but by that time they had dispersed, I was only able to get five horses at the most in any one picture.
My second trip to France was made in March. This time there was plenty of water in the marshes from spring rain, and I brought a 500mm f/4 telephoto. The horse wranglers forced the herd close together, and just as they started their forward motion I started shooting. The long lens enabled me to fill the frame, and I got what I wanted.
There were two major concerns, though. One was focus, as I’ve said. The other one was that once the herd of horses ran through the marsh one time, they became dirty. In fact, they were no longer white. They were muddy gray. Therefore, I had only one chance to get what I wanted at each sunset session.
Since I do not trust the AI servo mechanism in any camera, I had to use a different approach. Even though I used a 500mm lens, the horses were far enough away (about 300 feet) that I knew I had at least a small amount of depth of field at f/4. I used the largest aperture on the lens to get a fast shutter speed. As the horses were given the signal and they started churning the water, I focused (using a back button on my Canon 5D Mark II) and shot two or three frames on motor drive, and then I quickly refocused and repeated the burst of exposures. I did this several times — focus, shoot, refocus, shoot, etc. In just a few seconds the peak action was over. This picture is the best of the bunch.
- Tagged with:
- AI Servo
- autofocus
- focus techniques
- focus tracking
- predictive autofocus
Fujifilm's X-Pro1, now M Mount friendly
Olympus' Micro Four Thirds 75mm prime
Can you fix the focus on a blurry photo after the fact?
The birth of Mirrorless Cameras
The Joy Of Winning A Photo Contest
Choosing your first dSLR camera
New York City can be beautiful!
Choosing the Right Light Stand
Photojojo iPhone Telephoto Lens review — AudioCast
My week with Q
How To Become A Successful Photographer
"When the Wind Stopped" — poem with 4 photos
Creating The New Family Portrait
Tips for Textures
Cast aways - saving those photographic memories
One Man Show: My 25 Years With Digital Photography
Studio, Flash, & Available Light — Three Books Reviewed
Portrait styling: dangerous pairings
Adobe Photoshop CS6 Product Managers Interview Audiocast
A gift of flowers: unfold your senses
On Set of "Love & Robots" the Film
No-Brainer Setup For A Digital Photo Frame Exhibit - Part 3











San Diego 7 photo gallery — Just Be Love All Stay Cool
Planning “National Geographic” style photo travel
Wilderness Travel 1 Rainforests – Essential Gear
Backlighting Basics
What Moves You?
FIGURES IN MOTION: Decades of Evolving Personal Imagery in Photography, Part 7
Lomography Store, Austin, Texas — GALLERY
GALLERY — Up to $1,000 Reward for Cattle Rustlers
25% off on photography eBooks
eyePhone: The eBook for iPhone Photographers
Interview with Harold Davis — Closeup Maestro of Flowers & Water Drops
Interview with Steve Caplin — Photoshop Digital Artist, Commercial Illustrator, & Author
A Brief History Of Light & Photography: Part 3 of 3
A Brief History Of Light & Photography: Part 2 Of 3
Easy technique to select, edit and sequence keywords for web
How much should you charge for a photograph?








































Comments
I'm sorry-- you mention that AI *doesn't work* when the subject is moving towards or away from you. That it only works when it moves laterally across the frame (or up & down)?
Wouldn't that render AI Servo focus useless since typically focus only needs to be adjusted when the plane of the object perpendicular to the lens axis changes (i.e. moves forward or backward in relation to the lens)?
Or are you just saying that an object moving forward/backward in relation to the lens is better tracked by the AI Servo system when it is *also* moving laterally within the frame?
Thanks for the clarification,
Rishi
Post new comment