HDV to DVD in Linux

One way to create a standard definition DVD from an HDV tape is to set the firewire output to DV on the camcorder. This causes the camcorder to appear as a standard definition DV camcorder to the computer and the resulting workflow becomes the same as the DV Video Workflow in Linux. This obtains good results and allows you to keep the DV workflow that you might be already have. If you are interested in progressive frames, chroma keying or slow motion better quality can be obtained.

The workflow described here creates mjpeg Quicktime source files that use the 4:2:2 color space. We obtain high resolution slow motion, progressive and interlaced files simultaneously. Moreover, the progressive and slow motion video obtained is higher resolution than could be obtained from a DV source.

Starting with HDV input we use frame-rate doubling deinterlacing followed by rescaling to obtain 480p60 video. From this we can obtain

We use jpeg Quicktime for our intermediate video format because this is the most stable format to edit with Cinelerra. Note that the developers of Cinelerra originally wrote the Quicktime for Linux library. Cinelerra will edit jpeg avi almost as well.

Needed Software

This method relies on the following open source software.

Capture

First, capture the HDV source. Use the option -s512 to tell dvgrab to break the input into files with a size of 512K. This is important because when we double the frame rate to obtain slow motion we want to keep the files less than 2GB. 
$ mkdir capture
$ cd capture
$ dvgrab -fhdv -s512 -noavc src
Demultiplex each m2t file with ProjectX and then remultiplex with 
$ mplex -f3 src001.m2v src001.mp2 -r 28000 -b 800 -o src001.mpg
to avoid audio video desynchronization.

Another way of capturing is to download the whole mpeg2 transport stream as one m2t file with the option -s0 and then cut it into small enough pieces using ProjectX as part of the demultiplexing step. This has the advantage that you can be more selective in what scenes are contained in each file.

Creating the Quicktime Files

There is nothing to be gained in using a motion compensating deinterlacer, because the 540 lines of resolution in a single field of high-definition video is already more than the 480 lines in a standard definition frame. Therefore, we use the video filter -vf tfields=1:0,boxblur=1:1 in mplayer and select -vo yuv4mpeg as the video output device. The -nosound option avoids playing the sound track back in real time and greatly speeds things up. We split the yuv4mpeg stream after it is rescaled and simultaneously encode interlaced, progressive and slow motion video. This is done by the script hdvto480.

Note three things about the script. First, yuvfps tags the video stream at 29.97 fps before yuvscaler so that yuvscaler doesn't try to change the frame rate. Second, the yuv2lav encoder is forced into progressive mode with -I0 even for interlaced input so that Cinelerra can read the files. Third, if your version of mjpegtools doesn't have Quicktime support, the script may be modified by replacing .mov with .avi in the lines containing yuv2lav.

Now, create three Quicktime files srcYYYi.mov, srcYYYp.mov and srcYYYs.mov for each srcYYY.mpg file using the commands 

$ ./hdvto480 src001.mpg
$ ./hdvto480 src002.mpg
...
The output files are

For some reason vlc will not play the Quicktime files created above. These files play fine in mplayer and also load into Cinelerra without difficulties.

Editing In Cinelerra

You may now mix and match the .mov files on the Cinelerra time line. Cinelerra is quite happy editing jpeg Quicktime and will use smart copy for frames that are not changed when rendering the project to jpeg Quicktime output. Set the render options as

Smart copy also works when rendering your project using jpeg avi files.

Suppose that the file rendered by Cinelerra is named fin2jpeg.mov. Unfortunately, the frames encoded by Cinelerra use the 4:2:0 color space whereas the frames that were smart copied use 4:2:2. Therefore, fin2jpeg.mov contains a mixture of frames encoded with different color spaces. This file can be decoded by mplayer, mencoder, ffmpeg and lav2yuv; however, Cinelerra can't load it back onto the timeline nor can vlc play it.

Encoding to DVD

Convert the Quicktime or avi file to DVD compliant video stream using mjpegtools as described in DVD Compliant mpeg2 on Linux. Don't use ffmpeg or mencoder. If the video is interlaced, the interlacing mode needs to be explicitly specified with yuvcorrect when transcoding. This is because the original source was encoded by yuv2lav with the -I0 option.

Conclusions

If there is no need to produce a high-definition Blu-ray or HD-DVD disk, then it is it is easier to convert HDV to standard definition before editing. As a more flexible alternative to importing the HDV tape as DV footage from the camcorder, import the source as HDV and convert it to standard definition 4:2:2 jpeg Quicktime with yuv2lav. Note that that this Quicktime format appears to have more mosquito noise than DV.

An added benefit of this work flow is that you can go back and create a high-definition version of the edited video with no additional work. Simply substitute the original high definition source files into Cinelerra and rerender the project in high definition. To do this replace the srcYYYi.mov files with the srcYYY.mpg originals and the words "Quicktime for Linux" with "MPEG" in the xml Cinelerra project save file. If you used slow motion, you will need to create high definition slow motion files as described in Slow Motion Video in Linux and insert them into the xml file as well. Load the modified xml project file into Cinelerra, adjust the format settings, resize the tracks and render a high definition version of the project though a YUV4MPG Stream using one of the scripts contained in HD Video Workflow in Linux or H264 HD Video Workflow.

Evidence that mjpeg is not a good intermediate codec is presented by Eugenia Loli-Queru in Intermediate Codecs: the face-off where she states

Adam Wilt explains in his page DV vs MJPEG compression that Since we are editing on Linux, we follow the tradition of using mjpeg as our intermediate codec. Change the quality flag to -q95 or -q97 in the yuv2lav encoder to reduce the mosquito noise. It would be nice if Cinelerra could read and smart render lossless huffyuv encoded intermediate files.
Last Updated: Sun Mar 27 22:14:27 PDT 2011