Adding clock/timestamp overlay to camera feed

[sgbaird]

Not sure if the pi zero 2w's from #11 can handle both adding a timestamp and streaming in real time.

Related: https://www.reddit.com/r/VirtualYoutubers/comments/ifzqe7/i_made_a_live_timestamp_tool/

Lots of links here about adding it in real time with the Pi doing the processing:

https://www.google.com/search?q=add+timestamp+overlay+to+video+raspberry+pi

Cc @Jonathan-Woo

[sgbaird]

pre_callback seems like a nice option. Decided it's probably best to add on hardware itself, and if using picamera2 directly, probably not a huge deal in terms of processing overhead

[sgbaird]

@Jonathan-Woo could you give this a try and include a link to a stream with it?

[Jonathan-Woo]

Yes, pre_callback seems like it'll but the overhaul in 9b1a52b swapped out the picamera python implementation with the libcamera-vid. I tried digging through the chatgpt thread but couldn't find the rationale for this. @sgbaird could you fill me in on this please.

I haven't yet found a way to put the timestamp overlay with libcamera-vid.

[sgbaird]

Ah I forgot about that. It was related to empirical tests with stability:

#161 (comment)

#161 (comment)

At some point we just decided to stick with the "old script" which had demonstrated good reliability.

Given the timeline and the fact that we have such a backlog of videos to process that don't have timestamps already, maybe we punt on this one. If we're going to add the overlay in the other context anyway though, then maybe we can still add the clock there. However, processing the large number of existing videos should take priority.

[sgbaird]

Let's revert to picamera2 library, especially since we've now transitioned to cron-based 8 hour restarts. This will be important for us to be able to see how many frames are being lost (buffering on YouTube) even though the stream remains intact.

[Jonathan-Woo]

I've begun reverting to the picamera2 library and have added the overlay using pre_callback and the opencv library to add the timestamp to each frame based on this example:

[sgbaird]

I think I'm going to buy an analog clock and put it in the view 😄

[Jonathan-Woo]

I think the limiting factor is the video input thread_queue_size. I'm not sure what it refers to under the hood but it seems to fill up and not be ejected.

Increasing it to 16384 allows it to run for over 11 hours (https://www.youtube.com/live/iWH6ouveJeY) so with the 8 hour restarts it should work well.

Since it's not exposed by picamera2, I had to create a fork of it and increase the thread_queue_size manually (https://github.com/AccelerationConsortium/picamera2)

Though we should still test this with the restarts. Can we try swapping this script in on one of the camera setups please? I could also try but I'm unfamiliar with the youtube stream settings, auto restart code, and lambda stuff.

Created PR: #485

[sgbaird]

Yes, np. I can set it up. Please give me a permalink to the file I should run. Or can be link to file on the branch.

[Jonathan-Woo]

This file: https://github.com/AccelerationConsortium/ac-dev-lab/blob/213-adding-clocktimestamp-overlay-to-camera-feed/src/ac_training_lab/picam/device.py

But also you'd have to install the the picamera2 fork: https://github.com/AccelerationConsortium/picamera2

[sgbaird]

sudo apt update
sudo apt install -y python3-setuptools python3-pip python3-wheel python3-opencv

git clone https://github.com/AccelerationConsortium/picamera2.git
cd picamera2
python3 setup.py install --user