3D scanner accuracy is often tossed around loosely in spec-sheets and by sales reps. 1mm @ 25 feet, beam divergence, compensator accuracy... - it is very confusing and often does not reflect the true accuracy. Understanding the real world accuracy of a 3D scan is very important as most point cloud users expect point clouds to be perfectly accurate. Here we set out to test the real world accuracy of the Matterport 3D scanner.
What is Matterport
The Matterport 3D scanner is a low cost (~6,000 usd) tripod mounted scanner. The scanner is designed to be easily used by anyone and the main deliverable are 360 tours (very beautiful I might add). As a bonus a 3D model is created automatically from the scans and can be downloaded as a point cloud and imported into Revit/AutoCAD/PointCAB etc...
Matterport has focused on its virtual tour deliverable, however it can produce a point cloud and so it is an interesting potential cheap solution for those looking to dip their toe into 3D scanning or introduce point clouds on a project.
Matterport has focused on its virtual tour deliverable, however it can produce a point cloud and so it is an interesting potential cheap solution for those looking to dip their toe into 3D scanning or introduce point clouds on a project.
Method
To test the accuracy we scanned a 6,000 sqft floor with a Leica P20 and again with the matterport. The Leica P20 is an amazing instrument, (with a price tag of ~120k new, it should be) and for our purposes is perfectly accurate within a floor of this size. The comparison will highlight any shortcomings regarding accuracy of the Matterport 3D Scanner.
Results
We will test the accuracy in the horizontal (x/y), vertical (z) and levelness. Data is lined up manually in Revit for analysis.
RED is the Matterport data, BLUE is the Leica P20
RED is the Matterport data, BLUE is the Leica P20
Horizontal Accuracy
 |  |
Vertical Accuracy
Levelness
 |  |
Discussion
The horizontal accuracy is off ~5.5 inches over ~90 feet. 5.5 inches seems like a lot, but over 20 feet the accuracy should be around 1 inch, which seems reasonable.
The vertical accuracy shows a deviation of ~5 inches over a mere ~17 feet. This deviation is consistent across the entire floor. The reason for the large error is due to the dramatic falloff in accuracy with distance. The Matterport sensors cannot shoot directly up, and in order to reach the ceiling they measure at an angle and at the furthest extents of their range. Matterport suggests scanning closer together, or lifting the scanner on a high tripod to provide more accurate data on high ceilings.
The overall levelness of the data is excellent. No real difference between the Matterport and the p20. However, at the ends of the floor there is a "potato chip" warping effect. This is probably the same effect occurring on the ceiling: data far away from the scanner is less accurate (there were no scans directly near these areas).
One thing to note, we needed 2.5 times as many scans for the Matterport compared to the P20. This is due to the short range of the Matterport scanner.
Written by Thad Wester and Ben Yoder, Their views are their own and not necessarily those of WeWork.
The vertical accuracy shows a deviation of ~5 inches over a mere ~17 feet. This deviation is consistent across the entire floor. The reason for the large error is due to the dramatic falloff in accuracy with distance. The Matterport sensors cannot shoot directly up, and in order to reach the ceiling they measure at an angle and at the furthest extents of their range. Matterport suggests scanning closer together, or lifting the scanner on a high tripod to provide more accurate data on high ceilings.
The overall levelness of the data is excellent. No real difference between the Matterport and the p20. However, at the ends of the floor there is a "potato chip" warping effect. This is probably the same effect occurring on the ceiling: data far away from the scanner is less accurate (there were no scans directly near these areas).
One thing to note, we needed 2.5 times as many scans for the Matterport compared to the P20. This is due to the short range of the Matterport scanner.
Written by Thad Wester and Ben Yoder, Their views are their own and not necessarily those of WeWork.
