Unambiguous focus is the essence of optimized HD video. Maneuvering the lens-camera system, framing the sought-for imagery, altering focal length under director instruction and selecting the scene object (from within complex scenes) of sharpest focus are all multifaceted tasks that only the camera operator can precisely manage.
Within this multitasking challenge, a considerable number of factors conspire against the camera operator seeking to maintain sharp focus in an HD lens-camera system:
- Certain combinations of subject distance, lens focal length and lens aperture can further reduce depth of field.
- Fundamentally, HDTV has a shallower depth of field than SDTV.
- Defocusing sensitivity is much greater in an HDTV lens than its SDTV counterpart.
- Contemporary HDTV cameras use viewfinders that do not have HDTV resolution.
- The subject within a scene selected for sharpest focus may be in motion, posing unique tracking challenges.
The phase detection approach to auto-focus
Parallax occurs in panoramic photography if the camera and lens are not rotated around the entrance pupil of the lens and there is visible overlap between two adjacent images. Accordingly, to avoid parallax complexities, a through-the-lens (TTL) optical sampling system was adopted, which entails splitting off some of the light bundle sent by the lens to the primary camera imagers. This is augmented by a secondary image-registration, phase detection system that is used in most high-end SLR digital still image camera systems. However, the phase detection system for the HDTV lens had to be significantly refined beyond the capabilities of the digital SLR still imaging cameras. For the contemporary 2/3in long-zoom 100:1 HDTV field lens with a focal length of up to 930mm, the equivalent focal length translated to the 35mm image format is 3700mm.
Secondary image registration phase detection
Precise focus of the primary image plane (the HD camera's imagers) coincides with an exact focus of the two secondary images on their respective line array sensors. (See Figure 1.) When in focus, a hypothetical point is in precise point focus on the HD camera's three imagers, and that focus point is also precisely positioned in the center of the two line array sensors. The physical distance between these two centers is a specific reference known to the calculating algorithm.
Figure 2 outlines the action of the system for two distinct situations of primary lens defocus (Figures 2a and 2c) and shows one in focus (Figure 2b). Figure 2a shows the situation when the main lens focusing element is moved to the rear. The main image is clearly defocused on the camera's imagers and at the same time the two secondary images are defocused on their respective line array imagers.
What is of special note, however, is the fact that the latter two defocused images have also moved in position toward the outer extremities of the line arrays. The physical distance between them has increased to Dl. Knowing that distance has increased unambiguously informs the algorithm that the main lens element must be moved forward to begin the refocusing of the image. This known predetermination is a vital contribution to the overall speed of the system. Figure 2c depicts the shortening of the distance when the main lens focusing element is moved to the front.
As shown in Figure 3, the in-focus position produces two high-contrast secondary images at the precise reference distance Df. Even when grossly out of focus (in either direction), the secondary optical system is such that the detector can distinguish sufficient detail to allow determination of the distance between the two out of focus lower contrast secondary images.
Implementation of the auto-focus sensing system
The secondary imaging system is a specially built bonded arrangement of small lenses that sample the primary image plane in the manner shown in Figure 4. The two secondary imaging lenses project their separate optical replications of the primary object image onto the auto-focus (AF) sensor arrays.
Two lenses suitably positioned side by side with one another will create two separate images that are projected onto the two sensor arrays. It is important to note that these two optical subsystems are off-axis relative to the main HDTV lens optics. This creates an optical action that translates defocused images into horizontal movements of the two secondary images.
Of special note is that the secondary optical images consist of a small bundle of the total light rays that constitute the primary lens object image. Accordingly, their effective aperture is a higher F-number than that of the primary lens. Consequently, the effective depth of focus of the secondary image is considerably greater than that of the primary image. Thus, even when the primary image is considerably defocused, the longer depth of focus of the secondary lens system affords visibility of outlines that would be extremely blurred in the main larger image. The system has the powerful advantage in providing reliable detection information when the subject is way out of focus.
Array sensors for real-time detection of optical focus
To achieve the requisite detection speed for an HDTV lens providing full-motion imagery to the HDTV camera, a dramatic reduction is required in the readout time of the AF sensor array. In a new kind of AF phase detection sensor array, the individual sensors can be individually accessed and read out in less than a couple of milliseconds.
This is a major advantage over alternative AF systems that employ contrast detection, relying on the accumulation of a number of video frames that are examined by detection circuits seeking maximum contrast of the object image. The normal vertical cycle accumulation and readout entailed by a traditional area array imager takes 20ms for 50p HDTV (and 40ms for 50i). They also must pass through the actual focus point in order to verify that maximum contrast was indeed detected and then return to the point of sharpest focus. Thus, a system overshoot is inherent for verification. The phase detection system, on the other hand, provides multiple updates within a single video frame.
This is important when capturing moving finite distances within the period of a video frame. For example, a greyhound can move at up to 19m/s, cyclists at 130km/h (36m/s) and a racing horse at 65km/h (or 18m/s).
A field-programmable gate array (FPGA) microcircuit was developed to process the video from these sensors. High-speed processing of the detected data combines with high-speed computation within the dedicated CPU. This in turn controls a fast digital servo system that actuates the lens focusing elements. This processing is accomplished in a few milliseconds.
The special detection requirements of a motion picture HDTV camera
An HDTV lens married to a digital motion picture HD camera can operate at rates up to 60 full-frame pictures per second. This capture rate is six times faster than the contemporary digital SLR still image cameras. In particular, the HDTV long-zoom lens must contend with outside broadcast coverage of a wide array of sporting events — some of which can entail fast movement, such as in car racing, horse and dog racing, cycling, skiing, etc.
On a telephoto shot (typical of horse racing coverage), the depth of field is quite shallow. For example, at a lens setting of 930mm and an aperture setting of f4.7, the depth of field reduces to a mere 18m. A subject moving at 18m per scan can severely tax the camera operator challenged to maintain razor sharp focus throughout the run. Experienced operators can, of course, do this, but it does call for the highest visual attention, and there is a fatigue factor associated with this when operating an HDTV camera over long hours of event coverage.
In such instances, the assistance of a fast and reliable auto-focus system was considered a potential boon to the camera person. But, the system needs to reliably and accurately track the selected subject's movement and continually update the feedback loop that maintains the focus — all in a manner that is transparent to the viewer. That requires fast detection and rapid updating of the movement of the lens focusing elements — ideally in far less time than it takes to formulate one frame of HD video.
Auto focus in an HDTV studio lens
The HDTV studio environment poses its own special focusing challenges to the camera operator. Depth of field can become unusually short when imaging in the relatively low illuminations seen today in many contemporary studios. For example, for a 2/3in HD lens-camera system, depth of field is less than half a meter when the lens aperture is set to f2.8, focal length to 25mm and a subject is 5m from the lens front. Under such conditions, departure from a sharp focus can be encountered with a mere 15cm of subject movement. Thus, the unexpected movements of news anchors and talk show participants can easily cause a momentary defocus. In the critical environment of a home shopping show, unpredictable movements of show hosts and models (and especially the wares being portrayed) can impair HD lens focus.
Separately, the growing interest in integrated HD news studios has spawned the development of sophisticated automation and robotic systems that encompass the multifaceted operational dynamics of managing multiple video and audio sources, graphics, production switchers and effects, as well as studio cameras and lens controls. Automating the focus control of those lenses can be a major contribution to streamlining the automation of this highly complex production environment.
Summary
The advantages of the TTL phase detection system include:
- the ability to rapidly converge to a sharp focus even from a highly defocused situation;
- unambiguous predictive information that dictates which direction the focusing elements must be moved to restore accurate focus; and
- high-speed closed loop system that acquires multiple detection updates throughout a single video frame and facilitates accurate focus tracking of fast-moving subjects.
The overall accuracy and speed of the AF system has proven effective in many real-world field tests. The stress and fatigue associated with protracted HD coverage of long duration sporting events — where accurate focus of selected subjects in rapid motion is a high element of value-added production — has been alleviated. This offers an important new augmentation of creative HDTV shooting.
Larry Thorpe is the national marketing executive of the Canon Broadcast & Communications Division.
