best ffmpeg settings for cctv reencoding?

[eth]

Dear Doom9ers,

I have a video feed from Hikvision CCTV cameras which is of awesome quality even with low bitrate: open rtsp://doom9:doomdoom1@zoocafe.ath.c...g/Channels/101 in VLC (h265, 2048 kbps, 300 GOP, takes 5-10 seconds to load)

When I reencode it in OBS, the output quality is poor (I can see blocks and artefacts) even on "medium" tuning settings.

Somehow, the Hikvision company managed to figure out the holy grail of encoding even on their tiniest cctv chips (I took one of their cameras apart and was surprised by the low amount of electronics that they use).

Perhaps, there are special fine-tuning parameters for ffmpeg that I could use to get at least closer to that quality without overstressing the processor?

Since the feed is coming from a CCTV, the image is stable all the time (camera isn't shaking). Can I use any tuning parameters to account for this and devote more resources to encoding the actual movements of people/animals/nature within the frame instead of the image in general?

Thank you!

[benwaggoner]

Do you need to reencode it at all? If using HEVC isn't a blocker, couldn't you just mux the current stream as is?

[eth]

Remuxing it would work, but I have streams from several cameras + overlays, everything which I combine in OBS. Here is the live stream: https://www.youtube.com/watch?v=vKxpjUDQHBw

[benwaggoner]

Originally Posted by eth

Remuxing it would work, but I have streams from several cameras + overlays, everything which I combine in OBS. Here is the live stream: https://www.youtube.com/watch?v=vKxpjUDQHBw

With some fiddling of bits, you could convert each original HEVC stream into a tile in a bigger HEVC stream, without recompression.

Similar would be possible in H.264 with slices, but they'd have to be vertically stacked.

[rwill]

Originally Posted by benwaggoner

With some fiddling of bits, you could convert each original HEVC stream into a tile in a bigger HEVC stream, without recompression.

Similar would be possible in H.264 with slices, but they'd have to be vertically stacked.

No, an encoder producing a HEVC stream that is to be used as a tile in a larger bitstream needs to constrain certain aspects of the CUs at the picture borders or one get decoding errors. I dont think they took that use case into account when doing their cameras.