Currently if you pause LS with "lsause" and then step through a number of frames, e.g. "ls:step" or "ls:step 10", then you're not actually getting frame-by-frame results from the videofile, you're just getting a frame from where the video happened to be playing at the time.
This is not ideal for development, because you can't step through tracking logic frame-by-frame.
If LS is paused, then we need to pause the frame pipeline, and read frames from it individually using different code, following this example:
Here are test videos with timestamps for verifying behaviour: http://www.sync-one2.co.uk/support/test-files/