Python – Converting Pointcloud Data in UTM from TXT to LAS

laslaspyliblaspythonutm

I have text file with point cloud data in UTM format, say

sample=[632535.222, 4858488.332, 150.555]

I am currently using laspy, the parsing from txt is ready, and I plan to generate a new las file using the following code:

header = laspy.LasHeader(point_format=0, version="1.3")
outfile = laspy.LasData(header)
outfile.x = utm_east
outfile.y = utm_north
outfile.z = utm_z
outfile.intensity = raw_intensity
las.write(out_path)

I am getting the error message:

ValueError: invalid literal for int() with base 10: '632535.331'

which obviously because I am trying to store a float point into 32bit signed int as data format for LAS file.

Is there any way I can get around this with while retaining the precision of the result here? Even with scaling I think it goes beyond the maximum for 32bit int.

Best Answer

You must define a scale factor and an offset. I do not know python and laspy but according to the doc it should look like

header.offset = [632000, 4858000, 0]
header.scale = [0.001, 0.001, 0.001]

Related Solutions

[GIS] Converting between coordinate systems for LAS file using LASPY

You should check what the scale and offset are for your file. This can be done as follows:

van_taken.header.scale
van_taken.header.offset

This almost looks like an overflow error to me. The lower case x, y, and z properties need to re-scale and re-offset the coordinates to store it as an integer (which is how LAS files store them). To be honest, setting the values of the coordinates as scaled values is a bit of an anti-patern, because you lose control of how precision is lost.

Here's the code that handles your set operation, located in base.py:

    def set_x(self, X, scale = False):
        '''Wrapper for set_dimension("X", new_dimension)'''
        if not scale:
            self.set_dimension("X", X)
            return
        self.set_dimension("X", np.round((X -     self.header.offset[0])/self.header.scale[0]))
        return

If you can figure out what the unscaled integer representation of your converted data should be, that's probably a better way to store it (e.g., using the capital letter X, Y, and Z properties of the file).

If you're fine with the above approach to converting between the integer representation and the floating point representation, then I'd consider adjusting your scale to ensure that you don't end up with integers greater than four bytes in size.

If this isn't explicable via integer overflow due to scaling issues, then we definitely need to figure out what's going on. If it's an overflow issue, I'd be open to trying to guard against this case so long as it doesn't have too terrible a performance penalty.

Edit:

It looks like overflow is definitely the issue.

When you're assigning a scaled coordinate value into a LAS file, laspy needs to find some way of representing it as a four byte integer. Currently, it faithfully believes the information in the header. That is, it will subtract the appropriate offset (for the X, Y,or Z dimension) and divide by the appropriate scale (for the X, Y, or Z dimension). The result is then rounded to produce an integer.

Your file has an X scale of 1e-7, and an X offset of -83.11. Thus, to convert any new scaled value of x to its integer representation (which is what happens when you assign into the lower case 'x' property of your file), you need to add 83.11 and divide by 1e-7. For your first value, 269873.21570411, this results in a value of 2.699563e+12. The largest number you can store in four bytes is 2.14e9 for signed integers and 4.29e9 for unsigned.

Currently, laspy doesn't check for this mistake, resulting in an integer overflow. As I mentioned above, it's probably best to assign the integer values (to the capital X, Y, and Z properties) yourself to avoid any ambiguity.

As a quick fix, however, you can simply change the offset. The following ought to work:

van_taken.header.scale = [0.01,0.01,0.01]
van_taken.header.offset = [0,0,0]

You can increase the precision of your conversion by using a large offset and large scale. For example, if all of your scaled X coordinates are greater than 200000, you could use an X offset of 200000. Then, when a small scale like 1e-7 is used, the numbers it will be inflating will be smaller. That's something to play around with, keeping in mind the four byte limit.

In a lot of problems, and in a lot of computing environments, it's easy to gloss over the fact that floating point arithmetic is fundamentally not like real number arithmetic. Unfortunately, working with LAS files is not one of those cases.

Edit 2:

So can I change the scale value? Will this affect the data in any other ways?

Yes, the reason you can change the scale in this case is that you're supplying the scaled value. If you tell the LAS file that the scale is some particular value, that's the value it will use when re-scaling data. You wouldn't want to mess with the scale if you were reading existing LAS data.

Edit 3

Last question. So it seems that you were right about the scales. But I tried the .01 and it wasn't giving me the number. I then tried 1.000001 and it seemed closer, but they still aren't the same. Any tips for selecting the correct value, other than trial and error?

I don't think there's really a 'correct' value per-se, it's a trade off. No matter what you do, you're trying to store a (probably 8 byte) floating point vector as a vector of 4-byte integers multiplied by an 8 byte (double precision) floating point scale term and added to an 8 byte floating point offset term. That's not a lossless conversion in general, but it's what you have to do to store data in a standards compliant LAS file.

I would consider trying something like this (not tested):

converted_x = converted_csv[:,1]
converted_y = converted_csv[:,2]
converted_z = converted_csv[:,3]
xmin = np.floor(np.min(converted_x))
ymin = np.floor(np.min(converted_y))
zmin = np.floor(np.min(converted_z))

xmax = np.ceil(np.max(converted_x))
ymax = np.ceil(np.max(converted_y))
zmax = np.ceil(np.max(converted_z))

xrg = xmax - xmin
yrg = ymax - ymin
zrg = zmax - zmin

van_taken.header.offset = [xmin,ymin,zmin]
safety_factor = 2
maxval = 2e9
van_taken.header.scale = [safety_factor*xrg/maxval, safety_factor*yrg/maxval, safety_factor*zrg/maxval]    
van_taken.x = converted_x
van_taken.y = converted_y
van_taken.z = converted_z

[GIS] Visualizing LAS with matplotlib

This is a working solution with laspy:

import numpy as np
import laspy
from mpl_toolkits.mplot3d import Axes3D
import matplotlib.pyplot as plt

# reading las file and copy points
input_las = laspy.file.File("test.las", mode="r")
point_records = input_las.points.copy()

# getting scaling and offset parameters
las_scaleX = input_las.header.scale[0]
las_offsetX = input_las.header.offset[0]
las_scaleY = input_las.header.scale[1]
las_offsetY = input_las.header.offset[1]
las_scaleZ = input_las.header.scale[2]
las_offsetZ = input_las.header.offset[2]

# calculating coordinates
p_X = np.array((point_records['point']['X'] * las_scaleX) + las_offsetX)
p_Y = np.array((point_records['point']['Y'] * las_scaleY) + las_offsetY)
p_Z = np.array((point_records['point']['Z'] * las_scaleZ) + las_offsetZ)

# plotting points
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(p_X, p_Y, p_Z, c='r', marker='o')
plt.show()