Low cost scanning of architecture to CAD?
I just noticed the press release, "HOK, Leica Geosystems and Bricsys Announce "Scan-to-BIM" Initiative". I have been searching for a viable way to scan the warehouses where my company installs refrigeration chambers. It is not unusual to have to obtain measurements of a crowded warehouse. But, a system that is 10's of thousands is out of the question. So, I hope this initiative is seeking to get something viable for the small company. I will share some of my thoughts, in hopes that it gets some attention.
Ideally, such a scanner would be cheap, and easy enough to use, that we could even mail it to a distant customer, and talk him through setting it up to do a scan of his warehouse. Then, it would provide us a 3D scan of the facility. It would not have to be very detailed, in the 3D model. I don't need to measure the diameter of every piece of water pipe on the wall. But, if I can see an image that is detailed enough, as I do a virtual tour, I at least can see that there is a pipe there, and can ask for more information if needed.
There are several mature technologies to make measurements, so that such a device should be fairly cheap. The "Project Tango" proved to be too inaccurate for larger places, typical of business or industry. But it seemed to be designed for gaming, where they don't need as much accuracy.
In the U.S. you can get a laser distance meter for $40 that can provide +/- 1/16" accuracy at 100ft. I realize these are not doing continuous readings at a very fast rate, but a manufacturer could mount several sensors in a single unit, aimed at slightly different angles. Mount it on a tripod and let it slowly spin around the room to capture the walls and ceiling. And you don't need a zillion points of measurement to determine the dimensions of a wall. I realize that if a point is only every foot, it would not "see" things like windows, or water pipes on the walls. But, if an image were also taken, the user could see that they are there in a virtual tour. So, it could do a room scan in a minute or two. It may also have the intelligence to know when there is a sudden change in the direction of a wall, or a hole, that may be a corner or a doorway. So, it might automatically return to those areas to get a denser scan. The unit could be sold for less than $300, I am sure.
Other technologies are faster, such as the laser scanners that are now sitting on top of robotic floor vacuums., and Time-of-Flight cameras are also fairly affordable now, even if the pixel count is not very dense.
Concerning point clouds. There is on weakness in them that would affect scans of commercial buildings. Unless the point cloud density is smaller than the smallest item you want to "see", you would miss detecting things like trusses. Something like a laser with a fanned beam is a better approach for them. This technology is commonly used for scanning small things, like an 8" tall figurine. But, there is no reason it cannot be adapted to scan a ceiling to detect the trusses on a ceiling. Then, perhaps extract the elevation, and locations of the trusses, even if it does not try to model every detail of the truss.
For decades there have been university projects, where a college kid makes some very low cost way to scan a room. But, such things seem to never make it to the market. Hopefully, this initiative with BricsCAD involved, will produce something viable for the average user of BricsCAD.
For my fellow user, please speak up if you think I have hit onto something you are interested in seeing.
-Joe
Comments
-
Not sure if any of the above is Photogrammetry, as opposed to laser-radar type scanning?
In Photogrammetry a large no of still photos are taken from many different positions so any given object or point is seen in several perspectives; from that the system can calc ea point's position in 3D space. The result is dual - a 3D mesh-surface, and optionally a photorealistic image 'pasted' onto the surface.
In the SLAM variant, much used by estate agents etc, you just walk (or drone) through the space with a 360o camera firing continuously, a bit like a video, and the model is constructed simultaneously.
In the classic, high definition variant the photos are taken, uploaded and a big computer number-crunches for a while.Apart from that, professional surveyors use the expensive kit to produce a 3D model and/or 2D drawings, at remarkably low cost.
0 -
I didn't actually describe photogrammertry in my post. What I was describing with the laser is technically called "structured light scanning". This is where the laser is mounted near the camera, at a known distance and at a known angle. The laser might put out a single dot, or a line, or in the case of the Project Tango scanner, it is a whole lot of dots in a known pattern.
The process is like what a surveyor does using trigonometry. You have a known distance between the camera and the laser light source. Then you also know the angle that the laser light is aimed. The camera can determine the angle where it sees the laser hit the wall (or whatever) and given two angles and the length of one side, you can figure out the location of the point of that triangle. If the laser is putting out a line, that line forms a series of dots as dense as the resolution of the camera.
The photogrammetry you are talking about is not that much different. But, rather that the camera seeing the laser dot, it sees some object in the real world that it decides is a target. Great for finding corners, but not so good on a smooth wall without features, or ambiguous corners, or round columns.
SLAM (Simultaneous Location And Mapping) is another very interesting technology, where the camera figures out its position. But, it seems to be much harder to get an accurate scan from it. The idea of using a small drone is especially exciting, since it can fly around an area and see areas that would be hard to get to. But, with the camera platform in constant movement, it suffers a lot in terms of accuracy. Even the way a camera takes the image is not a simultaneous event, since the imagining system must get the data from the imaging chip in a sequential order. So, if the camera is in constant motion, that causes measurement errors.
I imagine a SLAM system can be made more accurate by giving the camera better targets to detect. It might be physical targets, but even just projecting dots from a laser mounted to a fixed base, would serve. SLAM is what the Project Tango does. But, when you try to scan a very large room, you do, and the software tries to match each scan to the next, all those errors start to add up. The beginning doesn't necessarily meet with the end. I've told that for a 30'x30' room, you might expect accuracy around +/- 6". That is too inaccurate to be useful.
-Joe
0 -
Thanks Joe - you obviouisly didn't need my simple lecture - in fact have explained some things in plain language that i've struggled to understand from online experts!
0 -
Cheap and accurate tend to be non-compatible in the same sentence, here is a fairly lengthy list of offers from $199 to $37,000
0 -
Steven, were you going to include a list?
Sometimes the cost is primarily a technology issue that later gets solved. Lasers used to be quite expensive, and now they are in the $1 store. Just think of what it would have cost to have a computer in your own home, in the year 1950.
The hardware for the scanning is cheap. Software is probably the more expensive part. But, again, this is stuff that has been college student projects for decades. The technology would seem to be quite mature. Though, perhaps things like automatically recognizing a pipe or I-beam, or door is not there. But, in my case, I don't see that this level of complexity is necessary for a useful product in my industry.
-Joe
0 -
You could take a look at the Leica BLK2GO. It's a wireless handheld imaging laser scanner. Believe it is intended mainly for internal spaces. There are more powerful devices available that are carried on a backpack.
Regards,
Jason Bourhill
BricsCAD V20 Ultimate
CAD Concepts0 -
@Joe Dunfee said:
Steven, were you going to include a list?It's the forum. It no longer highlights links. If you look at Steven's post (and mine), you will see he has included a link.
Regards,
Jason Bourhill
BricsCAD V20 Ultimate
CAD Concepts0 -
Thanks for the detailed instructions about how to discover the list that the forum has tried to keep hidden... is there a conspiracy?
But, sadly, the list is all about scanners designed to scan things much smaller than a warehouse. Though, if the distance between the light source and the camera increased, it should scale up its range by the same factor. So, separate the two by 2X, and you get 2X the range with the same precision. But, sub-milimeter precision is not needed for room scans.
And thanks Jason for the link to the Leica BLK2GO. But before anyone else goes bugging their sales people, it costs over $60,000 USD! Not quite what I was hoping to find.
-Joe
0 -
Agreed yes that list was presented on a site dedicated to 3D printing, but it does show a range of different methods used in 3D scanning at the lower end of the financial scale. Unfortunately for larger projects it does tend to be more limited and price is definitely in the higher bracket. This is a subject I follow closely (and yes I own a Tango enabled device Lenovo Phab2). The only real affordable method for anything of large scale at the moment is still photogrammetry which I have used on and off for about the last 20 years, the quality and accuracy depends on the software used and the camera equipment, my money is on 'Time of Flight' which is basically a camera sensor.
0 -
The challenge with the Time-of-Flight type of sensor is if they can see small things, like the trusses in the roof of most industrial facilities. Of course, the "flying spot' of many of these types of scanners can easily miss these types of targets, unless they are set to scan with a very dense number of points.
The advent of a Time-of-Flight camera, with its many pixels, has many advantages. But, I wonder if they are able to "see" things like roof trusses. They have, at best, 640 x 480 resolution. I wonder if an object is smaller than a pixel, perhaps enough light bounces back from that target area, that it detects it. For example, if the ToF camera is aimed at a chain-link fence, will it report the distance to the fence, or to the objects behind the fence. Or perhaps both?
-Joe
0 -
Steven, I didn't know you were into photogrammetry - would have talked to you about it if I'd known.
0 -
Tof Cameras don't use 'flying spots' they use a timed flash light source, and the sensor records the time it takes for that light to be recorded on each pixel of the sensor. Every pixel has image and time data recorded It uses a special flash that the sensors can then recognise the light frequency to record the time. I have been into photography since starting to develop my own pictures in 1997 and have owned many SLR cameras since then my first serious 'digital' camera was the Canon rebel (300D) and it was a major investment for just a 6MP camera, I remember at the time it was said a digital camera needed to get to about 20MP to start to be comparable to a 'normal' 35mm film camera, my current camera is 18MP and I am more than happy with the quality. The best I found at the moment is 150MP (but at $50,000 that's a bit beyond my budget). At 18MP each pixel is about 1mm at a distance of about 16 meters.
The accuracy of Tof sensors is about 1cm of depth. There are Tof camera's in practical applications with sensors 80x60 pixels, imagine what will be possible when those sensors become 5184x3456 pixels (18MP) and the depth accuracy is similarly improved.0 -
The better quality ToF cameras I see now are at 640 x 480 resolution.
I am really curious about what they do when a single pixel sees two different depths, such as at the corner of a wall, or an element of a truss.-Joe
0
