get started with point clouds - acfr/snark GitHub Wiki

Once you have snark installed, download datasets below (or use your own) and try the commands and pipelines of this tutorial.

The examples are for Linux. On Windows, it all should work in cygwin and many things in the command prompt.

Table of Contents datasets visualisation binary format interacting with an image

datasets

Below is a single Riegl lidar scan of the courtyard in front of Australian Centre for Field Robotics (ACFR). The ground and non-ground were algorithmically extracted and put in separate files:

Uncompress the files and take a look inside:

 gunzip rose.st.nonground.csv.gz
 head rose.st.nonground.csv

The files contain rgb-coloured cartesian points in csv format: x,y,z,r,g,b.

visualisation

view uncoloured points

 cat rose.st.nonground.csv | view-points
 cat rose.st.nonground.csv | view-points --colour=blue

view coloured points

 cat rose.st.nonground.csv | view-points --fields=x,y,z,r,g,b

by the way, view without unpacking

 zcat rose.st.nonground.csv.gz | view-points --fields=x,y,z,r,g,b

view both ground and non-ground

 cat rose.st.ground.csv rose.st.nonground.csv | view-points --fields=x,y,z,r,g,b

or better

 view-points --fields=x,y,z,r,g,b rose.st.ground.csv rose.st.nonground.csv

view both files with one fed on stdin

 cat rose.st.ground.csv | view-points --fields=x,y,z,r,g,b rose.st.nonground.csv

specify properties for stdin data ("-" conventionally stands for stdin)

 cat rose.st.ground.csv | view-points "-;colour=magenta" "rose.st.nonground.csv;fields=x,y,z,r,g,b"

colour only non-ground (fields=x,y,z is default)

 view-points rose.st.ground.csv "rose.st.nonground.csv;fields=x,y,z,r,g,b"

Notice that the csv file does not contain field names or any header or meta-information. It is the user who renders meaning to the data.

comma and snark utilities obviously will work even if you chop part of the file or process a stream of data (e.g. a live lidar feed), which by definition cannot have a header.

And you can reinterpret the meaning of the fields the way you like. For example, try to swap red and green colour channels on ground data:

 view-points "rose.st.ground.csv;fields=x,y,z,g,r,b" "rose.st.nonground.csv;fields=x,y,z,r,g,b"

or swap y and z in one of the files:

 view-points "rose.st.ground.csv;fields=x,z,y,r,g,b" "rose.st.nonground.csv;fields=x,y,z,r,g,b"

to be continued...

binary format

As you can see, loading such large files takes time, mostly because handling ASCII data is slow. Let us convert them into a simple binary format and see how it affects the performance.

You are unlikely to worry about the binary representation of your data, but here it is:

The binary format in comma and snark is the csv counterpart. It is simple and homogeneous, having several types, which have endianness of your system (i.e. mostly they are little-endian):

byte, char
signed and unsigned 2-, 4-, and 8-byte integer
4-byte float, 8-byte double
8-byte time in microseconds since Linux epoch (midnight of 1 January 1970)
fixed string

Since ASCII is human-readable and simple, I find it useful to start working with ASCII csv data first: get my pipelines working, debug problems, etc. Once everything works, I convert the whole pipeline to binary for performance (especially for heavy data streams, e.g. from lidars like Riegl or Velodyne).

convert a csv file into binary as 3 doubles followed by 3 unsigned bytes for colour

 cat rose.st.nonground.csv | csv-to-bin 3d,3ub > rose.st.nonground.bin
 cat rose.st.ground.csv | csv-to-bin 3d,3ub > rose.st.ground.bin

convert back to csv

 cat rose.st.ground.bin | csv-from-bin 3d,3ub > rose.st.ground.csv

or just take a look what is inside

 cat rose.st.ground.bin | csv-from-bin 3d,3ub | head
 cat rose.st.ground.bin | csv-from-bin 3d,3ub | less

view binary files

 view-points --fields=x,y,z,r,g,b --binary=3d,3ub rose.st.ground.bin rose.st.nonground.bin

view a binary and a csv file

 view-points --fields=x,y,z,r,g,b rose.st.ground.csv "rose.st.nonground.bin;binary=3d,3ub"

convert to binary and view on the fly

 cat rose.st.nonground.csv | csv-to-bin 3d,3ub | view-points --fields=x,y,z,r,g,b --binary=3d,3ub

interacting with an image

Left press and hold rotates the scene around the centre
Right press and hold translates the scene
Double left click changes the centre of the scene
Double right click outputs to stdout the coordinates of the clicked point

show points selected with a double right click using a named pipe

 rm -rf pipe && mkfifo pipe && cat pipe | view-points "rose.st.ground.csv;fields=x,y,z,r,g,b" "-;colour=sky;weight=10" --orthographic > pipe

publish a real time playback of georeferenced velodyne data on port 12345, visualise the data in real time and show points selected with a double right click

 cat velodyne-georeferenced.bin | csv-play --binary t,3d,ui | io-publish --size $( csv-size t,3d,ui ) -m 10000 tcp:12345
 rm -rf pipe && mkfifo pipe && cat pipe | view-points "tcp:localhost:12345;binary=t,3d,ui;fields=,x,y,z,block" "-;fields=x,y,z;colour=sky;weight=10" > pipe

similar to above but uses different colours for the shown points and adds labels next to the points indicating the click order

 cat velodyne-georeferenced.bin | csv-play --binary t,3d,ui | io-publish --size $( csv-size t,3d,ui ) -m 10000 tcp:12345
 rm -rf pipe && mkfifo pipe && cat pipe | view-points "tcp:localhost:12345;binary=t,3d,ui;fields=,x,y,z,block" "-;fields=x,y,z,id,label;weight=10" | csv-paste "-" line-number line-number > pipe

note: to make georeferenced data, run the following command on raw velodyne and novatel data

 cat novatel-raw.bin | novatel-to-csv --binary > novatel.bin
 velodyne_offset="0.0215,0.0026,-0.8822,-0.0014,0.0087,3.1143"
 cat velodyne-raw.bin | velodyne-to-csv -q --fields=t,x,y,z,scan --binary | points-frame --fields=t,x,y,z --binary=t,3d,ui --from $velodyne_offset --from "novatel.bin;binary=t,6d" | csv-from-bin t,3d,ui > velodyne-georeferenced.bin

to be continued...