I am using GStreamer to capture video from a USB webcam (Logitech C920) in H264, and I want to analyze the h264 frames before potentially decoding or streaming them to the net.
Based on different sources on the internet, I constructed a python2.7 script that allows me to get the frames into python, basically using the schematic gst-launch command:
gst-launch-1.0 v4l2src ! video/x-h264 ! h264parse ! appsink
However, I am stuck with trying to interpret the received buffers. I spent already quite some time trying to understand how python gstreamer works to get the meta api attached to the buffer, but in vain for now. If I understand correctly, if I get the meta api somehow attached to the buffer, I will get a structure that will allow me to access the different elements, and the information about the frame encoding. How can I do this? (without writing my own decoder for the frames)
Below is my current script, with some sample outputs:
from __future__ import absolute_import, division, print_function
import sys, os, pdb
from datetime import datetime
import gi
gi.require_version("Gst","1.0")
from gi.repository import Gst
Gst.init(None)
def appsink_new_buffer(sink, data):
sample = sink.emit("pull-sample")
buf = sample.get_buffer()
caps = sample.get_caps()
print("\nGot new buffer: {} Sample Info: {}\n".format(datetime.now(),sample.get_info()))
print("Buffer size: {} ".format(buf.get_size()))
print("Buffer n_memory: {} Presentation TS (PTS): {:.3f} s Decoding DTS: {:.3f} s Duration: {:.1f} ms".format(
buf.n_memory(), buf.pts/1e9, buf.dts/1e9, buf.duration/1e6))
st = caps.get_structure(0)
field_names = [st.nth_field_name(i) for i in range(st.n_fields())]
print("Caps {} n_fields: {} name: {}, format: {}, height: {}, width: {}".format(
i,st.n_fields(),st.get_name(), st.get_value("format"), st.get_value("height"), st.get_value("width")))
print(" all fields: {}".format(" ".join(field_names)))
for fname in field_names:
if fname not in ['pixel-aspect-ratio','framerate']: # cause error because Gst.FractionType not known
print(" {:20}: ".format(fname), st.get_value(fname))
#
# somehow, here one nees to get the Meta API to understand the buffer content and to do further processing
# of the encoded h264 frames.
# Q: does one buffer after the h264parse represent exactly one frame?
#
return Gst.FlowReturn.OK
def appsink_webcam_h264():
# adapted from https://gist.github.com/willpatera/7984486
source = Gst.ElementFactory.make("v4l2src", "source")
source.set_property("device", "/dev/video2")
caps = Gst.caps_from_string("video/x-h264, width=640,height=480,framerate=10/1")
capsfilter = Gst.ElementFactory.make("capsfilter", None)
capsfilter.set_property("caps", caps)
parse = Gst.ElementFactory.make("h264parse","h264parse")
sink = Gst.ElementFactory.make("appsink", "sink")
pipeline_elements = [source, capsfilter, parse, sink]
sink.set_property("max-buffers",20) # prevent the app to consume huge part of memory
sink.set_property('emit-signals',True) #tell sink to emit signals
sink.set_property('sync',False) #no sync to make decoding as fast as possible
sink.connect("new-sample", appsink_new_buffer, sink)
# Create an empty pipeline & add/link elements
pipeline = Gst.Pipeline.new("test-pipeline")
for elem in pipeline_elements:
pipeline.add(elem)
for i in range(len(pipeline_elements[:-1])):
if not Gst.Element.link(pipeline_elements[i], pipeline_elements[i+1]):
raise Exception("Elements {} and {} could not be linked.".format(
pipeline_elements[i], pipeline_elements[i+1]))
ret = pipeline.set_state(Gst.State.PLAYING)
# Wait until error or EOS
bus = pipeline.get_bus()
# Parse message
while True:
message = bus.timed_pop_filtered(10000, Gst.MessageType.ANY)
if message:
if message.type == Gst.MessageType.ERROR:
err, debug = message.parse_error()
print("Error received from element %s: %s" % (
message.src.get_name(), err))
print("Debugging information: %s" % debug)
break
elif message.type == Gst.MessageType.EOS:
print("End-Of-Stream reached.")
break
elif message.type == Gst.MessageType.STATE_CHANGED:
if isinstance(message.src, Gst.Pipeline):
old_state, new_state, pending_state = message.parse_state_changed()
print("Pipeline state changed from %s to %s." %
(old_state.value_nick, new_state.value_nick))
else:
print("Unexpected message received: ", message, message.type)
pipeline.set_state(Gst.State.NULL)
if __name__ == '__main__':
appsink_webcam_h264()
and below are some sample outputs of this script:
...
Got new buffer: 2016-01-09 01:41:52.091462 Sample Info: None
Buffer size: 9409
Buffer n_memory: 1 Presentation TS (PTS): 0.390 s Decoding DTS: 0.000 s Duration: 100.0 ms
Caps 8 n_fields: 9 name: video/x-h264, format: None, height: 480, width: 640
all fields: stream-format alignment width height pixel-aspect-ratio framerate parsed level profile
stream-format : byte-stream
alignment : au
width : 640
height : 480
parsed : True
level : 4
profile : constrained-baseline
Got new buffer: 2016-01-09 01:41:52.184990 Sample Info: None
Buffer size: 868
Buffer n_memory: 1 Presentation TS (PTS): 0.590 s Decoding DTS: 0.100 s Duration: 100.0 ms
Caps 8 n_fields: 9 name: video/x-h264, format: None, height: 480, width: 640
all fields: stream-format alignment width height pixel-aspect-ratio framerate parsed level profile
stream-format : byte-stream
alignment : au
width : 640
height : 480
parsed : True
level : 4
profile : constrained-baseline
Got new buffer: 2016-01-09 01:41:52.285425 Sample Info: None
Buffer size: 3202
...
I searched a lot but could not locate not one example how one could map the meta api to the buffers containing encoded video frames in python, and I assume that this shouldn't be that difficult, since the functionality seems to be provided.
Any suggestions?