6 Steps to Improve Measurement Accuracy in a Point Cloud Viewer | Prevent Misalignment with Coordinate Systems, Units, and Reference Points

Measurements using 3D point cloud data are increasingly being applied for as-built verification and design comparisons at civil engineering and construction sites. By measuring distances, areas, and volumes on a point cloud viewer, it has become possible to quickly verify post-construction shapes and to assess the amount of terrain change.

However, if point cloud data are handled incorrectly, unexpected deviations (errors) can occur in measurement results. For example, problems such as "dimensions that should match the design drawings do not match on the point cloud" or "when multiple point clouds are overlaid their positions are shifted and cannot be measured correctly" can occur. Many of these deviations are caused by data-setting issues such as inconsistencies in coordinate systems, units, or reference points. No matter how sophisticated the laser scanners or RTK-GNSS used to acquire high-precision point clouds are, if the reference coordinates or units are not aligned, large errors can result. In fact, there have been reports of cases where differences between the world geodetic system and local coordinate systems caused point cloud positions to be off by several meters (several ft), and where mixing up meters (ft), feet, and millimeters (in) caused scales to be significantly distorted.

In this article, we explain six steps to improve the measurement accuracy of point cloud viewers and prevent misalignment. We present, step by step, how to obtain reliable measurements from point cloud data by correctly standardizing coordinate systems and units and by utilizing reference points. Please use this guide as a reference to confidently carry out the precise measurements essential for on-site quality control within a point cloud viewer.

Table of Contents

• Step 1: Check and unify coordinate systems

• Step 2: Adjust the origin and height reference

• Step 3: Align units and scale

• Step 4: Correct and verify accuracy using reference points

• Step 5: Register multiple datasets (using ICP)

• Step 6: Final checks to improve measurement accuracy

• Conclusion: Recommendations for simplified surveying with LRTK

• FAQ (Frequently Asked Questions)

Step 1: Check and Unify Coordinate Systems

First, confirm the coordinate system. When comparing or measuring point cloud data against other data (such as design drawings or other point clouds), accurate measurements cannot be made if their coordinate systems do not match. A coordinate system is the reference that determines positions in the data, and it can range from global geographic coordinates (latitude and longitude, etc.) to arbitrary local coordinate systems. Common examples of coordinate system mismatches include the following cases.

• One dataset is in a global coordinate system (a geodetic datum such as WGS84), while the other is in a locally defined coordinate system set arbitrarily on site

• The survey coordinate system used when acquiring the point cloud differs from the coordinate system of the design drawings (for example: the survey results are in Japan Geodetic Datum 2011, while the design uses a custom local coordinate system)

• Errors in the setup of control point coordinates included in the data (known point coordinate values were entered incorrectly)

If coordinate systems differ, discrepancies will occur between positions on the point cloud and positions in the design. As a countermeasure, before loading data into point cloud viewers or point cloud processing software, unify the coordinate systems of each dataset. Specifically, transform the coordinates of the point cloud data as needed to match a common coordinate reference system. If the point cloud was obtained from on-site surveying, adopt the same reference system as that survey (for example, a public coordinate system or a plane rectangular coordinate system), and align the design data to the same standard. Conversely, you may convert the design drawings to local coordinates to match the point cloud, but it is important to align to one or the other.

If you are unsure which coordinate system to use, first check the official coordinate system specified for the project (construction coordinates or geodetic datum). For public surveys by the Geospatial Information Authority of Japan, this would be the Japan Geodetic Datum (JGD2011) or a specific zone of the plane rectangular coordinate system ◯. If you transform the point cloud to match that standard, the positional reference axes will align and major discrepancies should be resolved. Dedicated software functions for coordinate transformation (such as an affine transformation using three or more known points) are convenient, but if you know the corresponding points you can also do it manually. In any case, aligning the coordinate system as the "foundation" is the first step toward high-precision measurement.

Step 2: Adjustment of Origin and Elevation Reference

Next, adjust the origin (reference point) and elevation reference. Even if you think you have unified the coordinate system, if the data's origin position or elevation datum are offset, subtle discrepancies can remain in the vertical direction or on the horizontal plane. A typical case is when the elevation reference does not match. For example, if the point cloud elevations are recorded as elevations referenced to the Tokyo Bay mean sea level (T.P.), while the design data are referenced to the site's temporary benchmark (temporary BM), a vertical difference of several tens of centimeters (several tens of inches) may arise between them. Therefore, confirm and unify the reference surface for elevation as well.

Specifically, determine which vertical reference the point cloud heights are based on (mean sea level or ellipsoidal height, etc.) and convert them as necessary. If you want to align the reference to a temporary BM, translate the entire point cloud up or down in parallel to match the known BM height. In international projects, exercise extra caution because vertical datums (geoids) differ by country. Point cloud heights acquired with GNSS are often ellipsoidal heights and will not match local elevations as-is. Always apply a correction to the local official vertical datum (geoid separation).

Also, origin alignment is important. When point clouds are captured in multiple sessions, if the origin (the starting point of the coordinates) is inconsistent for each scan, the datasets will be slightly translated relative to one another. Decide on a single reference point and shift all point clouds so that that point has the same coordinate value. For example, use a clear common point, such as a site corner, as the reference and adjust the origin offset for each point cloud. If both height and the origin match, the datasets will mesh properly in three-dimensional space, and the foundation for measurement will be solid.

Step 3: Unit and scale consistency

Unit system (scale) verification is also essential for ensuring accuracy. If the length units of point-cloud data and design data differ, measurements will be displayed at the wrong scale. For example, if a drawing is created in millimeters but a point-cloud viewer interprets it as meters, a distance of 1,000 mm (39.37 in) will be interpreted as 1,000 m (3,280.8 ft), causing a discrepancy of 1,000 times. Conversely, if meters and feet are mistaken for each other, you can get a scale error of about 3.3 times, leading to large dimensional differences.

To prevent such unit mistakes, unify the data's unit system. Point cloud files often do not explicitly state unit information, so check the data source and the software settings. If the point cloud is meter-based (m, ft), set the viewer and CAD project settings to meters (m, ft) as well. In cases such as manufacturing data where millimeter units (mm, in) are used, either specify "unit=mm (in)" when importing, or scale the point cloud down by 0.001 to convert it to meters. Be especially careful when handling overseas data and check whether drawings or coordinate values use feet (ft). If unclear, contact the client or data provider to confirm the units and do not proceed based on assumptions.

After aligning the unit system, verify it once using known dimensions. For example, measure standard features in the point cloud such as the height of a standard door (approximately 2 m (6.6 ft)) or the roadway width (approximately 3 m (9.8 ft)) to confirm they yield realistic values. If you get values that are clearly off by an order of magnitude, the unit settings are likely incorrect. By correctly matching units and scale, the point cloud viewer's measuring feature will also perform at its intended accuracy.

Step 4: Accuracy Correction and Verification Using Control Points

Once the coordinate system and units are aligned, further improve the data's accuracy using control points. A control point is a point whose accurate coordinate values were surveyed on site. By utilizing control points, you can align point cloud data precisely with real-world dimensions and positions.

First, select several common feature points that appear in both the point cloud and the design data. These can be building corners, road intersections, corners of existing structures, or known points such as triangulation points or benchmarks. Survey these points on site in advance to obtain reliable coordinates and elevations. Next, determine the coordinates of the corresponding points in the point cloud, and based on the differences, apply a translation and rotation to the entire point cloud. Specifically, make adjustments such as "offsetting reference point A on the point cloud so that it matches A in the design coordinates." If necessary, correct the rotation angle as well and adjust using multiple reference points so that the overall error is minimized.

This kind of coordinate alignment using control points is easy when using dedicated software functions (such as a feature that automatically computes the transformation by specifying the coordinates of three or more corresponding points). Even when done manually, it is more efficient and accurate for aligning positions than matching coordinates point by point. The important thing is not to compromise on a visually "close enough" alignment. Always check numerically the displacement between control points, and verify whether the mean error and maximum error fall within the required accuracy. For example, if the offsets between all control points are within ± a few centimeters (± a few in), this can be considered sufficient accuracy for field surveying. If residual offsets exceed the tolerance, consider adding more control points or re-surveying. Correction and validation using control points greatly improves the reliability of measurements in a point-cloud viewer.

Step 5: Aligning Multiple Data Sets (Using ICP)

At large-scale sites, you may combine multiple point cloud datasets or overlay point clouds with design CAD data. In such cases, automatic alignment tools such as the ICP algorithm are particularly powerful. ICP (Iterative Closest Point) is an algorithm that aligns two point clouds by iteratively finding and optimally matching corresponding points, and it is effective for fine-tuning large misalignments.

Basically, if the coordinates and references are aligned through step 4, you should not see large errors even without relying on automatic alignment. However, when small misalignments remain or when comparing complex shapes that cannot be matched visually, it can be useful to leverage software. Some point cloud processing software or viewers have a function that, by simply inputting the two datasets you want to align, computes the optimal alignment using ICP. However, ICP will not converge correctly if the initial misalignment is too large. Therefore, manually pre-align them roughly (initial alignment) is required.

When you run ICP, small rotations and translations are automatically adjusted, allowing the datasets to fit together more precisely. In particular, when comparing large-scale datasets such as point clouds of an entire terrain and a design ground model, differences often become more uniformly small after ICP adjustment. You should not overtrust automatic tools, but if there are still noticeable misalignments even after aligning to reference points, try using it as an auxiliary tool. Point: do not forget to numerically verify the amount of misalignment at the end (Step 4). Don’t be complacent just because the alignment was performed automatically; by carefully checking with the human eye and calculations, you can ensure reliable measurement accuracy.

Step 6: Final checks to improve measurement accuracy

Finally, this section describes the final checks and tips when performing measurements in a point cloud viewer. The procedures up to this point have largely eliminated data misalignment, but there are still points to be careful about when actually using the measurement tools.

• Use of cross-sectional slices: Point clouds are large collections of 3D points, so depending on the measurement target you may not capture accurate dimensions unless you carefully choose the viewpoint and range. For example, when measuring width or thickness, slice the point cloud with a plane that passes through the object (cross-sectional view), and measure the distance between two points on that cross-section to ensure accuracy. By using a cross-sectional view, you avoid being misled by unnecessary foreground or background points and can measure the dimensions of the intended area with high precision.

• Handling noise and outliers: Point clouds may include measurement errors and noisy points. If there are points that are clearly floating or otherwise outlying, you may accidentally include them in distance measurements and obtain readings that overestimate the true distance. You can prevent such measurement errors by hiding anomalous values with the viewer’s filter function or by removing unnecessary points around the target before measuring.

• Duplicate data check: When combining multiple point clouds, duplicate point clouds can exist at the same location and appear denser. Taking distance measurements in this state can cause you to pick up shifts between data from different times. If necessary, perform integration/merge processing to create a single consolidated point cloud before measuring, which is ideal.

• Verification against known dimensions: Make it a habit to always measure a known dimension once before and after important measurements. For example, if the spacing of markings drawn on site is exactly 1 m (3.3 ft / 3 ft), measure that on the point cloud. If the result comes out around 1.000 m (3.281 ft), you can confirm there are no problems with coordinates, scale, or tool settings. Performing this kind of double-check on your actual measurement values also increases reliability.

If you take measurements with the above checkpoints in mind, you can maximize the accuracy of the values obtained in the point cloud viewer. To correctly assess on-site conditions, make sure to carry out the verification work carefully and remain vigilant until the end.

Conclusion: Toward Simplified Surveying with LRTK

If you put into practice the six procedures introduced so far, you can greatly improve the measurement accuracy of point cloud viewers and minimize data misalignment. However, conversely, that also means that achieving high-precision measurements requires a correspondingly large amount of verification and adjustment work. Isn't there a way to perform accurate point cloud measurements more easily? — A new solution that responds to that need has recently been attracting attention.

One of these is simple surveying using LRTK. LRTK is an innovative system that mounts a compact RTK-GNSS receiver on a smartphone and, while scanning point clouds with the smartphone's built-in LiDAR or camera, can assign high-precision positioning coordinates in real time. Previously, point clouds acquired with a smartphone alone had to be aligned afterward using control points, but by using LRTK you obtain point cloud data in the coordinate system that matches the design drawings the moment you scan on site. In other words, you can greatly reduce the tedious tasks of coordinate transformation and unit adjustment.

Furthermore, LRTK integrates with a cloud-based browser point-cloud viewer, allowing point clouds captured on site to be displayed instantly in 3D so users can measure distances and volumes and share data with stakeholders. Because high-precision GNSS positioning always assigns absolute coordinates to the point clouds, multiple scan datasets automatically overlap without discrepancies. This enables even non-experts to perform fast, highly accurate on-site measurements. If you are currently facing time or labor challenges with surveying or as-built management, it may be worth considering these latest simplified surveying tools. Leverage solutions like LRTK, which balance accuracy and efficiency, to smartly update measurement operations on site.

FAQ (Frequently Asked Questions)

Q: How reliable are the measurement results in a point cloud viewer? A: If the data are processed properly, distance and area measurements obtained in a point cloud viewer can achieve practical accuracy. With point clouds acquired by high-quality laser scanners, errors often fall below a few centimeters (below a few inches), and this is generally sufficient for typical civil engineering and construction as-built measurements. However, for control surveys where strict millimeter-level precision (≈0.04 in) is required, dedicated surveying instruments still have the edge. The key is managing accuracy according to the intended use, since point cloud measurement has the characteristic of being able to measure large areas at once while also detecting small errors. For that reason, pay attention to whether deviations exceed your acceptance criteria, and when necessary, verify using conventional methods in combination for reassurance.

Q: Even after aligning coordinates and checking units, there is still a discrepancy between the point cloud and the drawings. How should I deal with it? A: As possible causes, recheck whether there are any errors or oversights in the control point information. For example, if the coordinate values of the control points you used are themselves incorrect, or if the other dataset is using a different reference system, a slight offset can remain. Also, vertical discrepancies may be caused by a mismatch in elevation datum. In such cases, try conducting additional surveying of control points or calculate and apply a height correction. If it still doesn't match, you can also try the point cloud processing software's automatic registration function (such as ICP). However, do not leave everything to automatic processing; ultimately check the offset at the control points to confirm that the datasets are consistent.

Q: Is there a way to prevent misalignment during the point cloud acquisition stage? A: If you can acquire point clouds with minimal misalignment from the start, the burden of post-processing is greatly reduced. One effective method is to assign absolute coordinates during measurement using RTK-GNSS or similar. Recently, technologies have emerged that use devices such as LRTK, which combine a smartphone and high-precision GNSS, to acquire point clouds on site in real time with geodetic coordinates. This allows scanned point clouds to immediately align with map coordinates, eliminating the cumbersome coordinate transformation work that was previously necessary. Of course, with conventional methods you can also achieve precise alignment later by placing a sufficient number of targets or control points beforehand and capturing them during scanning. In any case, being mindful of reference points from the measurement stage dramatically improves the accuracy and efficiency of data alignment.