Skip to content

Latest commit

 

History

History
33 lines (26 loc) · 3.19 KB

README.md

File metadata and controls

33 lines (26 loc) · 3.19 KB
Raw

Pipeline Class

This is the parent class of the specific pipelines that will be built in the different applications. As a summary, the class loads a configuration file, creates elements/plugins and allows their use by adding and linking according to the applications.

This class allow us to create new applications easily. To create a new application we need to instantiate a new class NewPipeline and inherit our class Pipeline. Then, we just need to update two methods: create_pipeline and run_main_loop, according to specifications.

Initialization

The class is initialized through __init__ using a configuration file that is loaded from the parser. The default configuration file is ../configs/global_config.cfg and the first values loaded from it are the video sources.

Methods

The following methods allow us to create specific components of a given application. The explanation of each methods is below.

One of the most used method is _create_element(self, factory_name, name, detail=""), as the name suggests, creates a pipeline element using Gst.ElementFactory.make from the gi repository and also handles exceptions if an element cannot be created. In this way, the following pipeline elements can be created using this method.

  • _create_streammux(self) creates a batch of video frames before sending for AI inference.
  • _prepare_multistream(self, streammux) allows the creation multiple entries to the pipeline using the attributes self.sources, which was previously loaded in __init__ function. This method uses some functions of multistream.py, which prepare source bin. Note that you requiere to pass the streammux attribute for the creation of sinkpad.
  • _create_pgie(self, config_path) creates the primary inference.
  • _create_tiler(self) composites a 2D tile from batched buffers.
  • _create_tracker(self, config_path) creates a tracker plugin and enables multiple neural networks to reside within a single pipeline.
  • _create_sgie(self, name, config_path) allows to create secondary inferences.
  • _create_nvvidconv(self, name) performs video color format conversion.
  • _create_nvosd(self) draws bounding boxes, text and region of interest (ROI) polygons.
  • _create_capsfilter(self, filter_type) constructs a color channel filter for a filter_type: I420 or RGBA.
  • _create_nvv4l2h264enc(self) encodes RAW data in I420 or RGBA format to H264 or H265.
  • _create_rtppay(self) converts H264 or H265 encoded Payload to RTP packets (RFC 3984).
  • _create_h264parse(self) parses the incoming H264 stream.
  • _create_mux(self) merges streams (audio and video) into ISO MPEG-4 (.mp4) files.
  • _create_sink(self, sink_type=0) create a pipeline output according to a sink_type: RSTP, MP4 or Fake sink.
  • _create_tee(self) branches the incoming video streams in different source pads. Useful for Kafka.
  • _create_msgconv(self) converts metadata into event messages and consequently into a payload messages.
  • _create_msgbroker(self) sends payload messages to the server using a specified communication protocol.
  • set_probe(self, plugin, pad_type, function, plugin_name = "") allows to embed a probe function into the pipeline, indicating the place with the plugin and pad type.