Skip to content

Fork Discussion

This document explains the customizations made to the OCAP fork used in this project.

Repository

GitHub: https://github.com/shaundano/elephant-ocap

Why Use This Fork?

The standard OCAP installation may not work directly on non-GPU EC2 instances. This fork includes several modifications that enable OCAP to run on standard EC2 instances and adds additional features.

Key Modifications

1. Non-GPU Encoder Support

Overview: Swapped the NVIDIA encoder for a standard non-GPU encoder (x264enc). This allows OCAP to run on non-GPU EC2 instances (like t3.xlarge).

pipeline.py
screen_src |= "t. ! queue ! d3d11download ! videoconvert ! video/x-raw,format=NV12 ! x264enc tune=zerolatency speed-preset=ultrafast ! h264parse ! queue ! mux."

This line determines the encoder used for video recording.

2. Microphone Audio Capture

Overview: Added a second audio capture channel for microphone input. Allows recording both microphone and system audio as separate tracks.

pipeline.py
if record_mic:
    src |= ElementFactory.wasapi2src(loopback=False) >> "audioconvert ! avenc_aac ! queue ! mux."

The wasapi2src component is the same one used for recording system audio. The only difference between input and output audio is the loopback boolean parameter.

pipeline.py
record_mic: Annotated[bool, typer.Option(help="Whether to record microphone input as separate audio track")] = True,

The component is passed through setup and during the actual record function.

3. Process ID (PID) Automated Termination

Overview: Added PID file writing and graceful termination support. Windows doesn't have a graceful kill command like Unix systems. Sending a task kill to OCAP can corrupt files. This implementation allows graceful termination via SIGINT (Ctrl+C) from external processes.

Awesome - PID Termination Implementation

This might be the most based part of my project. I absolutely loved getting this to work. It is based on a Stack Overflow discussion. Read the stack overflow kids!

Writing the PID

recorder.py
pid_file = Path(r"C:\scripts\pid\ocap.pid")
pid_file.parent.mkdir(parents=True, exist_ok=True)
pid_file.write_text(str(os.getpid()))

Windows finds its PID and writes it to a static directory (C:\scripts\pid\).

Reading and Using the PID

stop-ocap.py
try:
    with open(PID_FILE, "r") as f:
        pid = int(f.read().strip())

The program reads ocap.pid and gets the process ID.

Sending SIGINT

stop-ocap.py
if not kernel32.GenerateConsoleCtrlEvent(CTRL_C_EVENT, 0):
    print(f"Failed to send Ctrl+C event. Error: {ctypes.get_last_error()}")
else:
    print("Successfully sent Ctrl+C signal.")
  • kernel32 is a Windows system library that provides low-level API functions
  • GenerateConsoleCtrlEvent sends the SIGINT signal
  • The SIGINT is defined as Ctrl+C, and 0 means it targets the entire group of processes attached to the console

Cleanup

recorder.py
finally:
    pid_file.unlink(missing_ok=True)

When OCAP wraps up, it deletes the PID file.

Congratulations! You've completed Phase 1. You should now have a fully functional setup with EC2, DCV, and OCAP running.