Canon EOS 5DS: iTR subject tracking at close subject distances

Here we look at the Canon EOS 5DS' ability to track a subject after initiating focus on it. The 5DS and 5DS R cameras have 150,000-pixel RGB+IR metering sensors that 'see' the scene to theoretically aid in subject recognition and tracking. Canon calls this system 'intelligent tracking and recognition', or iTR. We put it to the test.

We used a Canon 24mm f/1.4L prime for this test. The real-world relevance is that a camera with pinpoint subject tracking accuracy will continue to focus on the initially chosen subject (what was under the AF point when focus was initiated), no matter where the subject is within the frame. This potentially benefits focus of erratically moving subjects (kids, babies, pets, amongst many others), enables pinpoint accuracy for applications like shallow depth-of-field portraiture of moving subjects, frees the photographer from having to restrict composition by reframing to always keep the subject over the selected AF point in continuous AF, and can also help with the commonly used 'focus and recompose' technique, which otherwise runs the risk of focus shift from reframing and/or missed focus due to the subject moving by the time one has recomposed and the shutter is actuated.

We mostly initiate focus on the mannequin's eye, and then move the camera to simulate both recomposing and subject movement. The camera is unable to stick to the eye for the most part. Even holding the camera stationary often leads to jumping of AF points, and focus shifts (you can hear the focus motor working when it really shouldn't be), indicative of our observation that iTR can actually confuse the system and throw off focus sometimes with stationary subjects. We don't have much better luck when selecting other subjects either; however, when we select the red robot in the foreground, tracking appears to work better. This is probably due to either (1) its well isolated distance relative to other objects in the scene (and we're pretty sure Canon's tracking algorithm relies heavily on distance information, despite the presence of the 150k-pixel color metering sensor for color/pattern recognition), and/or (2) its quite obviously distinct color compared to other objects in the scene.

Overall, performance is not up to par with what competitors offer in implementations like Nikon's 3D tracking, and Sony's '4D' focus. We suspect the system relies too heavily on distance information to subject track, as this is how Canons pre-iTR worked. Briefly, the phase-detect AF module knows the distance of your subject initially when you half-press the shutter button, after which the system selects any AF point that reports a subject with similar distance (it's a bit more complicated, as the system also looks for contiguous movement across points, and any distance patterns indicating approaching or receding subjects). This sort of subject tracking struggles to offer pinpoint accuracy unless the subject is well isolated from other scene elements. Indeed, iTR subject tracking shows improved performance for subjects well isolated in depth, such as birds-in-flight, or a soccer player far out on a field, where there is a large disparity in subject distance between the subject and the background. We're perplexed as to why the subject tracking system appears to still rely so heavily on distance information. With iTR turned on, information from the color metering sensor does help stick to the right subject in some instances compared to a Canon camera without this hi-res metering sensor, but still not as well as a system that relies more heavily on actual pattern/scene recognition, such as the Nikon D810, which is accurate enough to stick to the eye of the mannequin in a similar test (   • NikonD810: 3D subject tracking at clo...  ).