Most Straight-Up Way to Retrive COMPLETE JPGs in OpenCV Environment

MOST of my JPGs come from SECURITY CAMS. They get refreshed, by
the cam, about once per second. Good enough.
But, problem, it's kinda pure CHANCE as to whether I get
a COMPLETEL jpg. The cams may update at any moment.
The RESULTS are too-often incomplete images.
Now software like Motion does find the proper
beginning/end of JPGs ... but it's kinda high
CPU. RTSP ... same, but even more CPU and
a LOT more latency. I have 8 cams now, complete
ring around the property, Motion will use TOO
much CPU. RTSP, a good minute or more behind
events unless you're using a major CPU. Do
NOT want to afford/use major CPU for this.
More like N-95/N-100 level. Abandoned the PIs,
just NOT quite enough.
This has bugged me for YEARS.
I like to use OpenCV (python) for this stuff.
Good lib. Really 'C' ... the Python is just
the execution infrastructure. No real advantage
to going all-'C' except obscurity.
Clearly though, OpenCV isn't entirely oriented
towards securing COMPLETE jpgs. It acts on
WHAT'S THERE at the moment. MIGHT be complete,
or maybe not.
As best I can tell, kinda I have to read the
jpgs byte-by-byte - accommodating/restarting
when the start bytes (too few IMHO) are detected -
and building the complete JPG in a buffer. Ugly.
DO-able, but ugly.
Fortunately my bandwidth/refresh is SLOW,
it's all low-end wi-fi at best. If I needed
25+ frames per - forget it. This means the
CPU I have has a *chance* at doing what
I want.
ANYway - does anybody here have better experience
with this particular kind of issue ... good links
at least ? Most of my searches don't turn up
anything really useful/do-able.
This crap is beyond 'theoretical', it involves
real external devices and such.