Loss of precision when importing LAS files with PDAL

Question

I check pdal info and see that bbox has 15 digit numbers:

"boundary": { "type": "Polygon", "coordinates": [ [ [ 1528634.540273188613355, 403421.710437842586543, 234.984 ], [ 1528634.540273188613355, 407979.483437842573039, 234.984 ], [ 1528834.540273188613355, 407979.483437842573039, 315.504 ], [ 1528834.540273188613355, 403421.710437842586543, 315.504 ], [ 1528634.540273188613355, 403421.710437842586543, 234.984 ] ] ] }

Then I imported this las file into pgpointcloud with this pipeline.json:

{
        "pipeline": [
            {
                "type":"readers.las",
                "filename":"%s"
            },
            {
                "type":"filters.chipper",
                "capacity":400
            },
            {
                "type": "writers.pgpointcloud",
                "connection":"%s",
                "table":"%s",
                "compression":"dimensional",
                "srid":"0",
            }
        ]
        }

The scheme in pointcloud_formats:

<pc:dimension>
  <pc:position>13</pc:position>
  <pc:size>8</pc:size>
  <pc:description>X coordinate</pc:description>
  <pc:name>X</pc:name>
  <pc:interpretation>double</pc:interpretation>
  <pc:active>true</pc:active>
 </pc:dimension>
 <pc:dimension>
  <pc:position>14</pc:position>
  <pc:size>8</pc:size>
  <pc:description>Y coordinate</pc:description>
  <pc:name>Y</pc:name>
  <pc:interpretation>double</pc:interpretation>
  <pc:active>true</pc:active>
 </pc:dimension>

And finally coordinates in table

with pts as(
    select PC_Explode(pa) as pt
    from "Blocks"
    where id = 100000
)
select 
     PC_Get(pt, 'X') as "x"
    ,PC_Get(pt, 'Y') as "y"
from pts
order by PC_Get(pt, 'Z');
-------------------------
x                   y
1531507.46827319    404533.469437843
1531507.47027319    404533.577437843
...

So in imported table precision is 8 digit numbers.

After I added both, the table and the las file in QGIS and found that the same point very very close but not exactly the same.

So did I lose precision? And how to prevent this?

The original numbers have 15 decimals and only 12 decimals are needed to express the diameter of a helium atom https://en.wikipedia.org/wiki/Picometre. — user30184, Mar 29 '22 at 15:41
@user30184 Well it's make sense but i just can't find a reason why my points move. And is it normal behavior or not. — Kliver Max, Mar 29 '22 at 15:52

score 1 · Accepted Answer · answered Mar 29 '22 at 15:43

1

You can preserve the precision by setting the scale factors in the point cloud writer, try adding these just below "type": "writers.pgpointcloud" :

        "scale_x":"0.000000000000001",
        "scale_y":"0.000000000000001",

From the docs:

Function docs: https://pdal.io/stages/writers.pgpointcloud.html

answered Mar 29 '22 at 15:43

Kartograaf

2,902
7
23

Any manipulations with scale_x/scale_y gives me error: Unable to convert double to int32 for packed DB output: X: (1.52877e+21). – Kliver Max Mar 29 '22 at 15:51
I think this has something to do with the output format you are using. Can you try changing to "type": "writers.las" and see if the problem persists? If not, one could assume the limitation is the the data type for your coordinate fields, which you can set manually in the schema as shown here: https://pgpointcloud.github.io/pointcloud/concepts/schemas.html – Kartograaf Mar 29 '22 at 16:14

score 0 · Answer 2 · answered Mar 29 '22 at 20:05

Bounding boxes are stored as double in a LAS files. But the point cloud coordinates are stored as integer and converted to double with a scale factor and an offset. It means that the point cloud coordinates are stored with a given accuracy (typically 2 decimals) but the bounding box could not conform to the standard and be stored with an irrelevant accuracy that does not match with the point cloud. I think your file does not conform to ASPRS standard but I may be wrong without seeing the file.

Anyway as mentioned in comments your accuracy is subatomic

Well, the precision is sub-ångström, but I doubt the accuracy is. — Vince, Mar 12 '24 at 18:25

score 0 · Answer 3 · answered Mar 12 '24 at 15:07

The problem is that PDAL is not conserving the previous offset, the default value for offset is 0 (source: PDAL las.writers page). This end up in it trying to save untranslated coordinates, which force it to drop some some significant values as the coordinates are too big. To correct it, you need to make PDAL offset the pointcloud when saving it.

If in command line, add :

--writers.las.offset_x=auto --writers.las.offset_y=auto --writers.las.offset_z=auto

if json pipeline, add to the writers.las :

"offset_x":"auto",
"offset_y":"auto",
"offset_z":"auto",

Loss of precision when importing LAS files with PDAL

3 Answers3

Linked