Andrew Wilson (a_wilson@xxxxxxx): > I'm using the code from > http://www.cse.msu.edu/~minutsil/linux/evi_grab/ > files > > grab.C > grab.H > grab_main.C > > pretty simple and straight forward, this isn't your code, it's setup for 400x400 capture, and it doesn't include your timing stuff. it was a little difficult for me to figure out what was going on. your code currently does not check how many buffers are used by the driver. your driver likely has like 2 gbuffers or 4 or something. basically, when you go to capture a frame, you say 'hey card, capture to this buffer right here', then you have to wait for the card to get told to write there, and this whole process wastes two frames of time, then the 33ms waiting for the capture to finish. i took the code and hacked it quickly to use double buffering, and now the call to SYNC only takes 33ms, and you correctly get the last frame captured. of course my hack is ridiculous and stupid but if you run the test you'll see the numbers are acceptable: http://vektor.theorem.ca/tmp/ Grabbing 640x480 images took 92 us for vidiocmcapture took 80012 us for vidiocsync( 0) took 96 us for vidiocmcapture ioctl VIDIOCSYNC: Invalid argument took 3519 us for vidiocsync( 1) took 105 us for vidiocmcapture took 84422 us for vidiocsync( 0) took 102 us for vidiocmcapture took 29091 us for vidiocsync( 1) took 101 us for vidiocmcapture took 29152 us for vidiocsync( 0) took 101 us for vidiocmcapture ignore te start part it's because it doesn't first free both of the frames. i highly recommend you rewrite the V4L capturing code in there, it's nasty. -Billy
