Take timestamp as close to shutter as possible

[fnoop]

The landing_target synchronised timestamp should be taken from as close to the shutter time as possible, before any image processing and pattern detection takes place. This is to more accurately correlate the inertial movement and attitude correction to the target image.

[fnoop]

c++ track_targets uses clock_gettime(CLOCK_MONOTONIC). python (2.7) vision_landing doesn't have inherent access to monotonic clock. Here's one way to do it: https://stackoverflow.com/a/1205762

[fnoop]

Python vision_landing uses:

    def current_timestamp(self):
        return int((datetime.utcnow() - datetime(1970, 1, 1)).total_seconds() * 1000000000);

This probably uses gettimeofday() to microsecond accuracy on Linux, although is CLOCK_REALTIME and thus subject to adjustments.

[fnoop]

So currently we can take the timestamp at great accuracy (clock_gettime monotonic) as soon as possible after the sensor frame is grabbed and before processing. This is output by track_targets and read by python, which will have to calculate where in the synced (with autopilot) clock object to insert this. As long as we get <1ms accuracy/correlation, this should be OK.

[fnoop]

Updated python code to use monotonic clock_gettime() clock and read new timestamp measure close to sensor grab. https://github.com/goodrobots/vision_landing/commit/cef02bd42b1ecf05706bbbecaf20bf4aec8ce12e