Ground removal lost resolution at long range

[tranqkhue]

Previous:

After:

[tranqkhue]

It seems by removing NaN value by https://github.com/tranqkhue/damn_3dlidar/blob/24560c268bcc8538eafc8c740c8eb8b8405bd8cc/test2.py#L60 the far range details are lost. This could be recovered by un-comment this line. However by not removing NaN values, there are two disadvantages:

• Cannot use Intel MKL to boost performance. Intel MKL when handle NaN value in function https://github.com/tranqkhue/damn_3dlidar/blob/24560c268bcc8538eafc8c740c8eb8b8405bd8cc/test2.py#L63 usually results in divergence error
• Regression algorithms fail to run LinearRegression does not accept missing values encoded as NaN natively.

[tranqkhue]

Solving the problem is temporarily by:

• Re-use OpenBLAS
• Removing NaN values is performed later