Accuracy Explained

The most-frequently asked question we get asked is about the accuracy of produced point clouds. So, how accurate is our data?

 

 

accuracyFirst of all, there's different kinds of accurate: we are mostly interested in what we'll call absolute and relative accuracy. Absolute accuracy describes how much the whole point cloud is offset in any direction, leading to a constant error in georeferencing. This error depends almost exclusively on a correctly configured GNSS reference station and thus, will not be discussed any further. Far more interesting is the relative accuracy, indicating how self-consistent the resulting point cloud is.

 

 

To answer this question, let's follow the history of origin of a point cloud's point: it is first scanned by a LIDAR sensor, then transformed into a global coordinate system using position and orientation solved by the GNSS/Inertial Navigation System.

 

 

 

 

Velodyne's HDL-32E and VLP-16 (Puck) LIDAR sensors are specified to measure distances with errors of less than 2cm. Riegl's VUX-1 lists a maximum error of less than 1cm. To be on the safe side, we'll assume an error of 150% the value specified by Velodyne, arriving at 3cm.

 

 

 

Because we mount LIDAR and IMU fixed to each other through a single flat aluminium part, we can consider them perfectly aligned and will not introduce an alignment offset (neither static, nor - worse - dynamic, e.g. due to vibrations), which might affect more fragile setups that place IMU and LIDAR sensors further apart.

 

 

The horizontal accuracy of the GNSS/Inertial position solution is listed as 1cm + 1ppm, where the latter figure indicates that for every kilometer of distance to the GNSS reference station (called the baseline), an additional millimeter of error is to be expected. Depending on the satellite constellation geometry, the vertical position error is usually estimated to be 150% of the horizontal error. Using a baseline of 1km as an example, we arrive at a positioning error of 1.983cm (1.1cm horizontal and 1.65cm vertical).

 

 

 

Summing up positioning errors from LIDAR and navigation system, our point is off by 3.983 cm.

 

 

However, the largest offset stems from orientation error in the GNSS/Inertial solution, which in turn is mostly determined by the accuracy of the IMU. For this reason, the AL3 and Ranger series use fiber-optic gyros and MEMS servo accelerometers. This is one of the most accurate commercially available IMUs (i.e., not affected by ITAR export controls), yielding 0.015 degree error in pitch/roll and 0.035 degree error in heading angles for real-time solutions - post-processing is usually even more accurate. To prevent losing alignment (common in environments with strong vibrations) and further enhance accuracy, the AL3 and Ranger series employ a dual-antenna solution (optional for Scout series).

 

 

Still, range and accuracy have always necessitated a compromise, because inaccuracies in orientation cause the point's error to grow linearly with it's distance from the LIDAR. This must also be noted when upgrading to scanners that offer longer ranges than the Velodynes (e.g. the Faro X330).

 

 

The image below details the range-constant positional error, as well as the additional, range-dependent offset due to orientation error.

 

 

Generally, errors in position and orientation are RMS (root mean square) values listed in the navigation system's specification. In practice, errors will not change rapidly within the given bounds, but drift slowly instead.

 

 

Even though all these numbers have their origins in thorough tests, they still are only numbers. Depending on satellite coverage and constellation geometry, vibrations, RTK baseline and choice of antennas, they are subject to change. We will gladly supply you with some sample data of ground and aerial surveys - please contact us!