diff options
Diffstat (limited to 'voctocore/lib/videomix.py')
-rw-r--r-- | voctocore/lib/videomix.py | 347 |
1 files changed, 347 insertions, 0 deletions
diff --git a/voctocore/lib/videomix.py b/voctocore/lib/videomix.py new file mode 100644 index 0000000..f98582a --- /dev/null +++ b/voctocore/lib/videomix.py @@ -0,0 +1,347 @@ +import sys, inspect, math +from pprint import pprint +from gi.repository import GLib, Gst +from lib.controlserver import controlServerEntrypoint + +class Videomix: + """mixing, streaming and encoding pipeline constuction and control""" + # size of the monitor-streams + # should be anamorphic PAL, beacuse we encode it to dv and send it to the mixer-gui + monitorSize = (1024, 576) + + previewbins = [] + + def __init__(self): + """initialize video mixing, streaming and encoding pipeline""" + # initialize an empty pipeline + self.pipeline = Gst.Pipeline() + + # create audio and video mixer + mixerbin = self.createMixer() + + # collection of video-sources to connect to the quadmix + quadmixSources = [] + + # create camera sources + for camberabin in self.createDummyCamSources(): + # link camerasource to audiomixer + camberabin.get_by_name('audio_src').link(self.pipeline.get_by_name('liveaudio')) + + # inject a ×2 distributor and link one end to the live-mixer + distributor = self.createDistributor(camberabin.get_by_name('video_src'), camberabin.get_name()) + distributor.get_by_name('a').link(self.pipeline.get_by_name('livevideo')) + + # collect the other end to add it later to the quadmix + quadmixSources.append(distributor.get_by_name('b')) + + # TODO: generate pause & slides with another generator here which only + # yields if the respective files are present and which only have a video-pad + + # add all video-sources to the quadmix-monitor-screen + self.addVideosToQuadmix(quadmixSources, self.pipeline.get_by_name('quadmix')) + + # initialize to known defaults + # TODO: make configurable + self.switchVideo(0) + self.switchAudio(0) + + Gst.debug_bin_to_dot_file(self.pipeline, Gst.DebugGraphDetails.ALL, 'test') + self.pipeline.set_state(Gst.State.PLAYING) + + def createMixer(self): + """create audio and video mixer""" + # create mixer-pipeline from string + mixerbin = Gst.parse_bin_from_description(""" + videomixer name=livevideo ! autovideosink + input-selector name=liveaudio ! autoaudiosink + + videotestsrc pattern="solid-color" foreground-color=0x808080 ! capsfilter name=filter ! videomixer name=quadmix ! autovideosink + """, False) + + # define caps for the videotestsrc which generates the background-color for the quadmix + bgcaps = Gst.Caps.new_empty_simple('video/x-raw') + bgcaps.set_value('width', round(self.monitorSize[0])) + bgcaps.set_value('height', round(self.monitorSize[1])) + mixerbin.get_by_name('filter').set_property('caps', bgcaps) + + # name the bin, add and return it + mixerbin.set_name('mixerbin') + self.pipeline.add(mixerbin) + return mixerbin + + def addVideosToQuadmix(self, videosources, quadmix): + """add all avaiable videosources to the quadmix""" + count = len(videosources) + + # coordinate of the cell where we place the next video + place = [0, 0] + + # number of cells in the quadmix-monitor + grid = [0, 0] + grid[0] = math.ceil(math.sqrt(count)) + grid[1] = math.ceil(count / grid[0]) + + # size of each cell in the quadmix-monitor + cellSize = ( + self.monitorSize[0] / grid[0], + self.monitorSize[1] / grid[1] + ) + + print("showing {} videosources in a {}×{} grid in a {}×{} px window, which gives cells of {}×{} px per videosource".format( + count, grid[0], grid[1], self.monitorSize[0], self.monitorSize[1], cellSize[0], cellSize[1])) + + # iterate over all video-sources + for idx, videosource in enumerate(videosources): + # generate a pipeline for this videosource which + # - scales the video to the request + # - remove n px of the video (n = 5 if the video is highlighted else 0) + # - add a colored border of n px of the video (n = 5 if the video is highlighted else 0) + # - overlay the index of the video as text in the top left corner + # - known & named output + previewbin = Gst.parse_bin_from_description(""" + videoscale name=in ! + capsfilter name=caps ! + videobox name=crop top=0 left=0 bottom=0 right=0 ! + videobox fill=red top=-0 left=-0 bottom=-0 right=-0 name=add ! + textoverlay color=0xFFFFFFFF halignment=left valignment=top xpad=10 ypad=5 font-desc="sans 35" name=text ! + identity name=out + """, False) + + # name the bin and add it + previewbin.set_name('previewbin-{}'.format(idx)) + self.pipeline.add(previewbin) + self.previewbins.append(previewbin) + + # set the overlay-text + previewbin.get_by_name('text').set_property('text', str(idx)) + + # query the video-source caps and extract its size + caps = videosource.get_static_pad('src').query_caps(None) + capsstruct = caps.get_structure(0) + srcSize = ( + capsstruct.get_int('width')[1], + capsstruct.get_int('height')[1], + ) + + # calculate the ideal scale factor and scale the sizes + f = max(srcSize[0] / cellSize[0], srcSize[1] / cellSize[1]) + scaleSize = ( + srcSize[0] / f, + srcSize[1] / f, + ) + + # calculate the top/left coordinate + coord = ( + place[0] * cellSize[0] + (cellSize[0] - scaleSize[0]) / 2, + place[1] * cellSize[1] + (cellSize[1] - scaleSize[1]) / 2, + ) + + print("placing videosource {} of size {}×{} scaled by {} to {}×{} in a cell {}×{} px cell ({}/{}) at position ({}/{})".format( + idx, srcSize[0], srcSize[1], f, scaleSize[0], scaleSize[1], cellSize[0], cellSize[1], place[0], place[1], coord[0], coord[1])) + + # link the videosource to the input of the preview-bin + videosource.link(previewbin.get_by_name('in')) + + # create and set the caps for the preview-scaler + scalecaps = Gst.Caps.new_empty_simple('video/x-raw') + scalecaps.set_value('width', round(scaleSize[0])) + scalecaps.set_value('height', round(scaleSize[1])) + previewbin.get_by_name('caps').set_property('caps', scalecaps) + + # request a pad from the quadmixer and configure x/y position + sinkpad = quadmix.get_request_pad('sink_%u') + sinkpad.set_property('xpos', round(coord[0])) + sinkpad.set_property('ypos', round(coord[1])) + + # link the output of the preview-bin to the mixer + previewbin.get_by_name('out').link(quadmix) + + # increment grid position + place[0] += 1 + if place[0] >= grid[0]: + place[1] += 1 + place[0] = 0 + + def createDistributor(self, videosource, name): + """create a simple ×2 distributor""" + distributor = Gst.parse_bin_from_description(""" + tee name=t + t. ! queue name=a + t. ! queue name=b + """, False) + + # set a name and add to pipeline + distributor.set_name('distributor({0})'.format(name)) + self.pipeline.add(distributor) + + # link input to the tee + videosource.link(distributor.get_by_name('t')) + return distributor + + def createDummyCamSources(self): + """create test-video-sources from files or urls""" + + # TODO make configurable + uris = ('file:///home/peter/122.mp4', 'file:///home/peter/10025.mp4',) + for idx, uri in enumerate(uris): + # create a bin for a simulated camera input + # force the input resolution to 1024x576 because that way the following elements + # in the pipeline cam know the size even if the file is not yet loaded. the quadmixer + # is not resize-capable + camberabin = Gst.parse_bin_from_description(""" + uridecodebin name=input + input. ! videoconvert ! videoscale ! videorate ! video/x-raw,width=1024,height=576,framerate=25/1 ! identity name=video_src + input. ! audioconvert name=audio_src + """, False) + + # set name and uri + camberabin.set_name('dummy-camberabin({0})'.format(uri)) + camberabin.get_by_name('input').set_property('uri', uri) + + # add to pipeline and pass the bin upstream + self.pipeline.add(camberabin) + yield camberabin + + + def createCamSources(self): + """create real-video-sources from the bmd-drivers""" + + # TODO make number of installed cams configurable + for cam in range(2): + # create a bin for camera input + camberabin = Gst.parse_bin_from_description(""" + decklinksrc name=input input=sdi input-mode=1080p25 + input. ! videoconvert ! videoscale ! videorate ! video/x-raw,width=1920,height=1080,framerate=25/1 ! identity name=video_src + input. ! audioconvert name=audio_src + """, False) + + # set name and subdevice + camberabin.set_name('camberabin({0})'.format(cam)) + camberabin.get_by_name('input').set_property('subdevice', cam) + + # add to pipeline and pass the bin upstream + self.pipeline.add(camberabin) + yield camberabin + + def iteratorHelper(self, it): + while True: + result, value = it.next() + if result == Gst.IteratorResult.DONE: + break + + if result != Gst.IteratorResult.OK: + raise IteratorError(result) + + yield value + + def previewBorderHelper(self, previewbin, enabled, color = 'red'): + crop = previewbin.get_by_name('crop') + add = previewbin.get_by_name('add') + add.set_property('fill', color) + for side in ('top', 'left', 'right', 'bottom'): + crop.set_property(side, 5 if enabled else 0) + add.set_property(side, -5 if enabled else 0) + + + ### below are access-methods for the ControlServer + + @controlServerEntrypoint + def numAudioSources(self): + """return number of available audio sources""" + liveaudio = self.pipeline.get_by_name('liveaudio') + return str(len(list(self.iteratorHelper(liveaudio.iterate_sink_pads())))) + + + @controlServerEntrypoint + def switchAudio(self, audiosource): + """switch audio to the selected audio""" + liveaudio = self.pipeline.get_by_name('liveaudio') + pad = liveaudio.get_static_pad('sink_{}'.format(audiosource)) + if pad is None: + return 'unknown audio-source: {}'.format(audiosource) + + liveaudio.set_property('active-pad', pad) + return True + + + @controlServerEntrypoint + def numVideoSources(self): + """return number of available video sources""" + livevideo = self.pipeline.get_by_name('livevideo') + return str(len(list(self.iteratorHelper(livevideo.iterate_sink_pads())))) + + + @controlServerEntrypoint + def switchVideo(self, videosource): + """switch audio to the selected video""" + livevideo = self.pipeline.get_by_name('livevideo') + pad = livevideo.get_static_pad('sink_{}'.format(videosource)) + previewbin = self.pipeline.get_by_name('previewbin-{}'.format(videosource)) + + if pad is None or previewbin is None: + return 'unknown video-source: {}'.format(videosource) + + self.previewBorderHelper(previewbin, True, 'green') + for iterbin in self.previewbins: + if previewbin != iterbin: + self.previewBorderHelper(iterbin, False) + + pad.set_property('alpha', 1) + for iterpad in self.iteratorHelper(livevideo.iterate_sink_pads()): + if pad != iterpad: + #self.previewBorderHelper(iterpad, 0) + iterpad.set_property('alpha', 0) + + + @controlServerEntrypoint + def fadeVideo(self, videosource): + """fade video to the selected video""" + raise NotImplementedError("fade command is not implemented yet") + + + @controlServerEntrypoint + def setPipVideo(self, videosource): + """switch video-source in the PIP to the selected video""" + raise NotImplementedError("pip commands are not implemented yet") + + + @controlServerEntrypoint + def fadePipVideo(self, videosource): + """fade video-source in the PIP to the selected video""" + raise NotImplementedError("pip commands are not implemented yet") + + + class PipPlacements: + """enumeration of possible PIP-Placements""" + TopLeft, TopRight, BottomLeft, BottomRight = range(4) + + + @controlServerEntrypoint + def setPipPlacement(self, placement): + """place PIP in the selected position""" + assert(isinstance(placement, PipPlacements)) + raise NotImplementedError("pip commands are not implemented yet") + + + @controlServerEntrypoint + def setPipStatus(self, enabled): + """show or hide PIP""" + raise NotImplementedError("pip commands are not implemented yet") + + + @controlServerEntrypoint + def fadePipStatus(self, enabled): + """fade PIP in our out""" + raise NotImplementedError("pip commands are not implemented yet") + + + class StreamContents: + """enumeration of possible PIP-Placements""" + Live, Pause, NoStream = range(3) + + + @controlServerEntrypoint + def selectStreamContent(self, content): + """switch the livestream-content between selected mixer output, pause-image or nostream-imag""" + assert(isinstance(content, StreamContents)) + raise NotImplementedError("pause/nostream switching is not implemented yet") |