Make nan values detection case insensitive

[VictorLamoine]

NaN value detection was case sensitive, it is now case insensitive.
The boost header for this function is already included.
#1003

[taketwo]

👍

[jspricke]

We had this discussion back in the beginning of PCL and removed the feature of different nan styles, because it slowed down io quite a lot. That's why I wrote the pcl_pcd_convert_NaN_nan tool back then. I didn't test your implementation and afair the old one was different. Still we should check the speed implications of this change.

[VictorLamoine]

PCL latest trunk compiled in Release mode
I used pcl_pcd2ply program (only the loading part) to check the loading time of pcd files.

The point cloud has been generated with MeshLab;
Filters > Create new mesh layer > Noisy isosurface > 256

The file contains no NaN values.

PCL latest trunk :

$ pcl_pcd2ply noisy_isocloud_ascii.pcd /dev/null.ply
> Loading noisy_isocloud_ascii.pcd [done, 11939 ms : 1768939 points]
> Loading noisy_isocloud_ascii.pcd [done, 12137 ms : 1768939 points]
> Loading noisy_isocloud_ascii.pcd [done, 12139 ms : 1768939 points]

With my pull request :

> Loading noisy_isocloud_ascii.pcd [done, 12219 ms : 1768939 points]
> Loading noisy_isocloud_ascii.pcd [done, 12121 ms : 1768939 points]
> Loading noisy_isocloud_ascii.pcd [done, 12231 ms : 1768939 points]

There is a 0.97% loss of performance from these results.

Then I used a simple program to changes values to nan in the point cloud. (see 2nd commit)

PCL latest trunk :

$ pcl_pcd2ply noisy_isocloudNaN_ascii.pcd /dev/null.ply
> Loading noisy_isocloudNaN_ascii.pcd [done, 13641 ms : 1768939 points]
> Loading noisy_isocloudNaN_ascii.pcd [done, 13709 ms : 1768939 points]
> Loading noisy_isocloudNaN_ascii.pcd [done, 13623 ms : 1768939 points]

With my pull request :

> Loading noisy_isocloud_nan_ascii.pcd [done, 13655 ms : 1768939 points]
> Loading noisy_isocloud_nan_ascii.pcd [done, 13769 ms : 1768939 points]
> Loading noisy_isocloud_nan_ascii.pcd [done, 14141 ms : 1768939 points]

There is a 1.42% loss of performance from these results.

I think the loss of performance is negligible.

[jspricke]

Thanks a lot Victor!