Avconv snapshot from videostream not complete

[yfands]

Hello,

I'm having some trouble with getting a snapshot from a videostream. Sometimes the snapshot looks good, but most of the time the picture is 'not complete'.
The attached picture is one of the worst, most of the time the picuture is complete for 75% and the bottom part blurt.

I'm using a (dedicated) Rapsberry pi 3 as platform with Raspbian GNU/Linux 8 (jessie).

A misecu (chinese camera) resolution 1920x1080 with a 3518 processor, no onvif, no snapshop capability, only rtsp stream.

and I am using the following avconv command (removed IP, user and password)

sudo avconv -rtsp_transport tcp -i rtsp://IP_PORT:554/user=_password=_channel=1_stream=1.sdp?real_stream –rtp-caching=100 -f image2 -vframes 1 -pix_fmt yuvj420p "/var/www/html/snapshot/snapshot_oprit.jpg"

If I use the parameter stream=0 the picture is ok but the resolution lower and out of proportion. But I need the hires snapshot.
I'm no expert in avconv and really no idea where to begin to tweek.

Oh, maybe not relevant but the Pi and camera are connected via WIFI.

Anybody got any ideas ?

Regards Frank.

[LigH.de]

IMHO this is clearly the result of data corruption. But the reason is not so obvious for me, I lack of knowledge in this specific case.

Well possible that the data transfer stalled (e.g. because the video bitrate is higher than the network capacity in this specific scene, or the transcoding is not suitable for realtime processing), or – you mentioned WiFi – there was a radio interference burst. Or it just recorded one TCP packet at a maximum size (maybe 64 KB?), but the content of the video frame is larger than only one TCP packet. Someone with more knowledge may be able to pick the best from my wild guessing.

If possible in any way, I would recommend to convert the stream separately from its recording. You should get at least one whole GOP stored before you attempt to convert it.