On Fri, Jun 26, 2009 at 7:55 AM, Georg Martius<georg.mart...@web.de> wrote: > Hi, > > On Thursday 25 June 2009, Francesco Romani wrote: >> On Mon, 2009-06-15 at 11:31 -0500, Carl Karsten wrote: >> > So, anyone interested in hacking that together for me? Or have some >> > ideas... etc. >> >> Just an Ack/KeepAlive. >> >> I'm very pressed those days, but I have strong interest in that kind of >> things (and of course the Fedora selinux bug plus the usual business). >> >> +++ >> >> Given the current and future architecture of transcode, and given the >> scope and the purpose of the project, I think the implementation could >> be "just" a filter that evaluates the differences in a given sequence of >> frames and log (to a file) the frame id/timestamp on which a given >> motion measure exceed a given threshold. > > A similar code is in the stabilization plugin. It detects motion of subsequent > frames and writes it into a file. > Since it uses small measurement fields one can also detect small > moving objects, however the code has to be modified a bit. Just a though.... > > This kind of motion detection would be better than a simple difference image, > because brightness changes and so on are no problem, but off course it is > slower.
I am OK with slower - the last part of my work flow is a script that runs for an hour or so, so I let it run over night so there is plenty of time. Another use for this would be to cut out duplicated frames (caused by camera going gooffy or something.) right now the picture freezes for a second or so, sometimes 10 seconds. if there is no audio, then there is no point in making the viewer watch that period of time when there was no video, so drop that section. There will be a hickup as it jumps over the part that dropped, but that's better than wasting time watching the still image. -- Carl K
