add processor_capture_holder

Add a processor which sole purpose is associate a capture of any type to a key-frame creation.
The processor keeps a time-windowed list of CaptureBasePtr from any sensor, on the keyFrameCallback it associates the closest capture it holds to the key-frame.

This can be convenient in situation such as :
A laser sensor is used as the main odometer through ICP. Its CaptureLaser are transformed into CaptureMotion within the processor getIncomingCaptureMotion function. Thus there is no laser anymore to build any map representation from the lasers. This processor aims at 'keeping' the scan for later map construction.

@jsola The newly added CaptureBuffer class may need a bit of polishing in the capture retrieval by stamp.

@artivis should I merge this?

I am recently advocating for having CaptureMotion act as an abstract class, therefore requiring everybody to derive it. Should this eliminate the need for your solution? Should we discuss it further? Should I merge this anyway?

J

CaptureMotion as an abstract class for all captures ?? Yeah we may need to discuss.
The work here is nothing but a buffer of CaptureBasePtr and a Processor that query for a capture given a key-frame stamp and associate the found CaptureBasePtr to the frame.
I think this can be merged shall it evolve later on.

Not for all captures, but for all captures that require motion integration:

• Odom3D
• Odom2D
• IMU
• maybe others.

Not sure about motion via LIDAR with ICP, or via Vision and trifocal tensor, etc, since in these motion processors there is no need to integrate small motions.

Merging in 3, 2, ....

However, look there are warnings:

/Users/jsola/dev/wolf/src/./capture_buffer.h:161:33: warning: returning
      reference to local temporary object [-Wreturn-stack-address]
  return (!container_.empty())? container_.front()->getTimeStamp() :
                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/Users/jsola/dev/wolf/src/./capture_buffer.h:167:33: warning: returning
      reference to local temporary object [-Wreturn-stack-address]
  return (!container_.empty())? container_.back()->getTimeStamp() :
                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Halted merge. Please check warnings.

merged

mentioned in commit 46e6670d

The warning comes from the fact I return the InvalidStamp object introduced earlier in case the buffer is empty.
I'll have a look on how to fix this.