Home Made Lidar Church Scan
HTML-код
- Опубликовано: 22 июн 2018
- This is a render of the data produced by my hobby build Lidar after a scan of a church. The full build log video will be ready soon. The Lidar was built using a Garmin Lite V3, an STM32 micro-controller pair, and a Warco lathe and mill. The rendering is in Unity3D. Mechanical design was in Fusion 360, PCB design in Eagle, and PCB manufacture by PCBway.
Extremely impressive again David! This project really encompasses so many engineering areas and as such there are very few people who can pull everything off. Super excited to see the build video and interested in what you are going to do with it.
Thanks “CogWheel” - that’s really appreciated. The build video is actually ready and all I need to do is press the publish button. The only sticking point is sorting out some background music that I like. I’ve already got verbal permission to use it from the record label, so it’s just a case of making sure that it’s formalized with RUclips.
Depending on your file type, you may be able to view your data using a point cloud viewer such as Realworks Viewer instead of a game engine drawing spheres. Well done on building your own 3D scanner! Quite impressive!
Congratulations! A fascinating project and excellent results, especially so for a home build
Very impressed.!
Amazing!!!
David would you mind sharing how you made the LiDAR Scanner?
Impressive David! This could be improved and use to scan the forest.
Quite a large point cloud.
Can we use tf mini lidar(tof) instead of lidar lite v3 ?
Fascinating. How long does the actual scan take and how many points does it capture (not asking you to count manually!).
Thanks Ian. It was about 0.75 million points for the church, which took about 10 minutes to scan. A small room (in my first lidar video) only needs a couple of minutes to get the same “point density” over the surfaces.
Would it be possible to scan even finer?
Great project David! Whats your background?
Thanks . Physics by education, software for the day job, machining for fun!
What would happen if you took multiple scans from different positions in the church and then married all the different scans together?
I think it would be a really interesting thing to do. The difficult part is finding some way of aligning the scans. That’s easy if the Lidar has positional sensing inside it i.e. accelerometers, GPS, inertial, compass bearing, and the like. This all ends up being something called Lidar SLAM. But, my project has none of this, so it’s either by eye or some no doubt rather complicated maths.
You could go the lowtech route.
1. define a convenient baseline in the object you want to scan. Every scan is oriented to that.
2. Define two spots with a known distance on that baseline. These are the measurement points to locate all the other scanning spots.
3. Every additional spot you want to start a scan from get a distance measurement from that spot.
Afterward you solve the triangles to get the absolute distance and angle from that baseline first scan. With these information overlaying the different 3D-Images should be easy.
If the ground you work on is not flat you need three points of reference, but in my experience a laser lever will help you work on the same plane.
I measured the dimensions of a basement this way.
But you have to remember if the measurements was "left" or "right" of the baseline.
You could go a more technical approach. Get foam balls with a know diameter and positioning them in the room in a way that at least three of them are visible in all scans. Write a program that searches the dot-cloud for a round shapes of the known size. Combine all scans by aligning them by the known ball-positions. I imagine that this can be a little bit process intensive but should work without additional sensor equipment.
On the stepper motor controlling the sensor elevation I’ve got a hall probe and magnet so that it can find the home position, and it has to do that otherwise you can’t locate the motor. On the stepper motor controlling azimuth the zero degrees point is where ever it happens to be when it’s powered up. I figured that was OK, because even if you know where you are with reference to the rest of the Lidar, the Lidar itself is still plonked down in some random orientation. That of course doesn’t help with your first suggestion as some sort of rotation will always be needed between the two point clouds, but it’s also easy enough to fix. I really like the second idea, which is obviously harder to code but computationally it’s probably OK because you only have to do it once and not every rendered frame.
I personally would find the second approach more interesting. Simply because I've never done it. :-)
I think a problem could be to identify the "anchor-balls" reliably. You need at least four points on the ball to get it identified. Ideally much more. Three points are needed to define the center of the anchor-ball and at least one more to verify that it is the ball. But if to little points hit you will get false positive which will trow your aligning off. So a value how well defined the position of a "anchor-ball" is will be needed. You then start with the best anchors (most "hit-points") and then add less reliable ones if needed to define the position and orientation of the scan. With more than two scans aligned you can eliminate false hits, because they will only appear in one scan. The additional points of the scans will make the anchor-point more reliable. And with more and more scans added you can eliminate more and more false anchors.
But it's a brute force approach. You have to check every point in the 3D-cloud with the surrounding points in all your scans. And then you have to do the same with the known centers of the anchor-balls.
And it gets harder. The points of the surface of the balls will fluctuating in distance and position a little so variation has to be allowed in the checking. And more variance will appear the further the ball is from the scanner.
Also the balls have to be quite large to make it reliable on distance, but not so large that they obstruct to much of the field of vision. And the position in the space should be as random as possible to make false alignments impossible.
So far my thoughts on that...
I will preface my post with the fact I am just a land surveyor that happens to get the chance to use a Leica C10 Lasers scanner from time to time and I Just learnt from using it and doing how to use it it and how to produce models from it.
What you are talking about doing is how we use commercial scanners produce their large models. You position targets plates or as what Boelwerkr mentioned in their post spheres around the site so you can see as many as possible from the different scan positions. if in the case of doing work for surveying ( as is my use) we ill have some of these points coordinated with a position in relation to some grid system that our work is fixed to, but if it is just to produce a stand alone model of an area the relative position ( distance between 2 of them and the height difference ( say measure from the floor to the center of the balls ) and use that to scale the model in a cad package , and then from the original scaled model bring in a each of the next ones and scale them to the now known points in the already scaled model.
even if you dont have positioned targets in the scan, measuring the length between the columns and use that to scale the original scan then you can pick points in the separate scans that are the same and fit them to the original
this is a very crude description of how commercial automated unit do their stuff.
David what you have done is amazing,
Any chances you would share this lidar file? thx
Holy sh... I mean Lidar!