It's been a good few weeks for DSLRs. Just after Nikon's D90 became the first ever to jump into the sweet, sweet waters of the HD video pool, Canon did a gigantic cannonball today with the EOS 5D Mk II, upping the game to full 1080p captures at 30fps. The question that all of this good news brings up is: Why now? Why haven't the DSLRs we've been using for years ever been able to grab video clips like their cheaper pocket-cam brethren? Let's take a look at the roadblocks that have stood in the way of the DSLR video revolution, and why Canon, Nikon and others are only just now starting to Bigfoot right over them.Processing: The biggest challenge to overcome is also the most vague and nebulous, and is the one part of digital cameras most of us never think about. Processing is each manufacturer's secret sauce (that's why we never hear a MHz number or any other specs). It's where a huge portion of the engineering dollars go to-and where most of the patents come from. The data dumped out by a DSLR's large sensor is a lot to swallow-orders of magnitude more than what even the baddest compact digicam can process into video. So for instance, according to Canon's long-time camera tech guru Chuck Westfall, the 5D Mk II's Digic IV processor must take the 21MP, full-frame sensor data and downsample it by 10x to 1920 x 1080 and compress it into MPEG4 encoding-and it has to do that 30 times each second. That's a lot of CPU power, and Canon and Nikon both only just got there, with Digic IV and Expeed, respectively. Auto Focus: If you've used live view on a Nikon or Canon DSLR, you know that auto focus happens differently. That's because with the mirror flipped up and the shutter open, the channel is cleared to allow light from the lens to stream onto the main image sensor, bypassing the separate AF sensor used for still images. It's the same for a DSLR in video mode-on the 5D Mk II, the camera uses a separate contrast-based system to assign focus points (or it can also attempt facial recognition using contrast, both of which put even more strain on our good friend the processor), and it takes several seconds for focus to shift if it needs to (in some modes you must assign the new focus point manually using the camera's jog dial). While you may not notice or care about shifting auto focus during videos on a compact camera, when a DSLR's more responsive and sensitive depth-of-field is factored in, focus is more important. The majors are just now locking all of this down enough to the point of usability-and they still have a ways to go. Of course, you can always focus manually, but try getting a major manufacturer to put that dusty sentence in a PR brochure. Sony and Olympus have used an innovative two-sensor setup to provide live view without the AF problems. However, Olympus has gone away from that model and now only uses one sensor, presumably to cut down on cost and complexity. Nobody has used the second image sensor to dump the live view feed to video, probably because the output would not be of usable quality. Sensors: Everything changes when instead of the fraction of a second of exposure for a still image, a sensor has to sustain constant operation to grab a video. When it's capturing light continuously a sensor heats up, and heat = noise = shitty looking images. Today's DSLR CMOS sensors (which handle heat and noise better, generally, than CCDs) are just now getting to the levels of low power consumption and efficiency to not turn into little mini-griddles when recording a video. Canon has only produced one CCD camera in its history (the first, the 1D), and Nikon has been CMOS on the top end-but there's a reason the mid-range video-capable D90 has moved to CMOS from its D80 predeccesor's CCD, and why most manufacturers are heading in that direction. Olympus uses what they call LiveMOS (or NMOS) and may very well be implementing video recording soon on its E-series cameras. Image Quality: When you buy a DSLR, you want everything that comes out of it to be of substantially better quality than what you could get with the compact digital it's likely replacing. That's why the first DSLR movie modes we've seen from the biggies are all touting HD quality-if they had really wanted to, someone could have found a way to squeeze video out of a DSLR before now-but for all of the reasons above, it wouldn't have looked much (if any) better than what compacts have been spewing directly to YouTube for years. And while the majors want you to love your new "prosumer" status, they're also quite happy to have you continue shopping for a compact camera to back it up. But now, the more interesting overlap is not DSLRs vs. point and shoots, it's DSLRs and digital camcorders. We'll have a Giz Explains on "should I even consider buying a digical camcorder again" ready when that becomes an actual reality. Special thanks to Chuck Westfall from Canon for helping us out. Something you still wanna know? Send any questions about touching, feeling or screening to firstname.lastname@example.org, with "Giz Explains" in the subject line.
DPReview has some info about the format the 5D-MII shoots: h.264 at 38mbps, full 1080/30p, PCM uncompressed audio, in the MOV container. Unfortunately, that 30p is either 30.00 fps or a variable of [30.xxx] fps, instead of NTSC's standard 29.97. This means that an editor like Sony Vegas will resample by default, and this creates extreme ghosting in the finale export (unless you manually disable resampling in all your clips in the timeline — something that not all video editors support).
Canon should have done the right thing: selectable frame rates at EXACTLY 29.97, 25.00, 24.00 and 23.976.
To me, as an final step editor (rather than a shooter), this is my main complain. It makes editing and final quality much harder than it should be.