Robot Projects > Rolling

Point cloud anyone !!! maybe 3D scanner



Its scanning, the stealthy Dual Time of Flight wiring was a soldering pig to execute (or should I say "Exterminate").

Data is flowing.... at momo a bit strange 200 Steps = 360° (until microstep instigated giving theoretical 1/16 steps ermm if my maths is correct 3200 Steps for 360° however at the price of speed)

In the current set up it is possible to gimbal the lasers both on the X and the Y orbits (i.e. putting the robot on its side and using one of the drive wheels as one of the Gimbals. (it could be redesigned as a pure standalone 3D scanner ...I think I am talking myself into another project.... darn it.... \ö/ )

A quick PointCloud of first few points after reveals :-

Converting Polar to Cartesian coords.

x = TOF_Scan × cos( TOF_Angle° ) : y = TOF_Scan × sin( TOF_Angle° )

Current 3D mount block will take 4 modules...

Now I need to find best way to - work with/display - the information......

.... ideas greatly welcome....

In the past, I have used two servos mounted on each other so I can do a rotation along a 3D sphere and do a scan. To make distant measurements I used a Lidar Lite, though it's rather slow, at least the old version.

It created a nice 3d point cloud of my room. For further processing I planned on using ROS or the PCL (point cloud library), however this is only well supported under Linux, so I am still waiting for the Windows version of ROS to get better.

Since MS is working on it, there might be hope now.

Your point cloud viewer/processor can be found here (for future reference):

I used a point cloud viewer that looks pretty similar to the one in your screenshot. Can't find the name anymore. It also supports mesh reconstruction, and allows playing around with it a bit in quite intuitive ways (rotation, scaling, coloring, etc.). Quite useful.

Edit: Found it, while searching for yours! It's called MeshLab:

I also played around V-REP to test a virtual 3d scanner, to see what I was doing wrong/right, since part of my scan looked a bit distorted. The reason was that 3d rotation around two angles probably does not form a perfect sphere, because of the relative positioning of the two servos/pivot points. You would need to measure physical offsets from the ideal rotation points and consider that.

I was going to calibrate it with a real 3d box, then use a box of the same measurements in V-REP. But I haven't continued since then. The 3d scan is really slow, and I have no 3d box (of even shape here). Bummer.

Also, I will probably use a Intel RealSense going further. They seem reasonable in size and use now. And much faster than doing a mechanical 3d scan. Could be used to guide a robot through a room by making quick distance measurements in various directions based on the "instantly" (in one frame) captured point cloud.


[0] Message Index

Go to full version