Five Causes of Point Cloud Misalignment｜Countermeasures for Reference Points, Coordinate Systems, and Errors

Table of Contents

• Cause 1: Errors in reference points and setup mistakes

• Cause 2: Mismatch of coordinate systems

• Cause 3: Insufficient overlap or lack of feature points

• Cause 4: Instrument errors and effects of the measurement environment

• Cause 5: Data processing mistakes and configuration errors

• Summary

Point cloud data alignment (registration) is an extremely important process in construction, civil engineering, and surveying. When integrating multiple scan datasets or overlaying a collected as-built point cloud with design drawings or 3D models, if positions are not accurately matched, errors will arise in shape comparisons and dimensional measurements. For example, if point clouds are misaligned when comparing as-built conditions to the design, it can undermine the reliability of quality checks and earthwork volume calculations, and can lead to construction mistakes and rework. Therefore, performing point cloud alignment correctly and minimizing discrepancies contributes to improved site efficiency and ensures accuracy. Accurate point cloud alignment can also be regarded as a foundational technology that supports digital construction.

However, in practice, problems such as "when multiple point clouds are combined, some parts do not fit together" and "when point clouds are overlaid with design data, their positions are significantly off" often occur. Why do these kinds of discrepancies happen? In this article, we discuss the five main causes of point cloud misalignment and explain countermeasures for each cause. The content is widely applicable to measurement data from terrestrial laser scanners and drone-mounted LiDAR. By correctly understanding the causes and responding appropriately, you will be able to make high-precision use of point cloud data. At the end of the article, we also touch on a new high-precision positioning method that fundamentally resolves these mismatch issues.

Cause 1: Reference point errors and setup mistakes

When aligning a point cloud to a known coordinate system, mishandling reference points (control points) can cause large positional shifts. For example, if you mistype a digit when entering the coordinates of a known point installed on site, or if you reference the wrong control point, positions that should match will no longer align. Also, care is needed when there is only one control point. If you force alignment using only a single point, you cannot correct angular differences or scale discrepancies across the site, and the error may grow with distance. Especially in large-area surveys, insufficient control points can result in positional discrepancies of several centimeters to several tens of centimeters (a few inches to a few dozen inches) at distant locations. Furthermore, in the vertical direction, if the elevation value of a control point is set incorrectly, the entire acquired point cloud will be shifted up or down by that amount. Even an input error of only a few centimeters (a few inches) can become a non-negligible error when judged against strict as-built quality criteria.

As a countermeasure, when performing alignment using control points, first carefully verify the coordinate values of known points in advance to prevent input errors. Have multiple people cross-check and use a checklist to minimize human error. Also, using as many control points as possible is effective for improving accuracy. Use at least two known points to correct planar rotation (orientation), and, if possible, use three or more points to adjust for scale differences as well. For example, on a site where the coordinate axis is slightly rotated from true north, aligning to a single point can produce misfits that cannot be fully corrected, but correcting with multiple points will suppress local discrepancies. After alignment, it is essential to compare with other independent control points and verify the offsets between corresponding points. Check whether the average error and the maximum error are within acceptable ranges, and if there are locations that fall outside the standard, consider additional control-point surveying or readjustment.

Cause 2: Coordinate System Mismatch

If the coordinate systems of point cloud data and the comparison target (design drawings or other point clouds) do not match, significant positional discrepancies will occur. Coordinate system mismatch is one of the most serious causes of point cloud alignment problems. For example, point clouds acquired by laser scanners are often in a local coordinate system with the instrument as the origin, whereas design drawings and CAD models may have been created in a public coordinate system based on known site control points, or in a project-specific local coordinate system. If one is in the global geodetic system (latitude/longitude or plane rectangular coordinates) and the other is in an arbitrary site-specific origin coordinate, the values for the same point can appear offset by tens of meters or more. For example, a point labeled (120.00 m (393.70 ft), 50.00 m (164.04 ft)) on a drawing would, when expressed in the national reference coordinate system, become values near (200000 m (656168 ft), 50000 m (164042 ft)), and thus appear to be completely different coordinates at first glance. Also, because the origins differ by region in Japan’s plane rectangular coordinate system, if the zone (system) number is set incorrectly the entire dataset can be off by tens to hundreds of meters. Furthermore, attention is needed when the azimuth (rotation angle) of the coordinate axes differs. For example, if a site-specific coordinate axis is tilted 10° from true north, at a point 1 km (3,280.8 ft) away a lateral displacement of approximately 170 m (557.7 ft) occurs. Thus, aligning just a single point is not enough because the fit worsens with distance; alignment that corrects for rotation and scale is indispensable. In addition, when old drawings were created in the old geodetic datum (Tokyo Datum) and newly acquired point clouds are in the global geodetic system (JGD2011, etc.), there will be a constant offset of about 300 m (984.3 ft) in the east–west direction and about 150 m (492.1 ft) in the north–south direction.

Differences in coordinate reference frames also pose a problem in the vertical direction. Heights obtained from point cloud surveys are usually the values called "ellipsoidal height" when positioned by GNSS. On the other hand, the heights used in civil engineering design are generally "elevation (orthometric height)" referenced to mean sea level. Ellipsoidal height and elevation differ by about 30–40 m (98.4–131.2 ft) around Japan, and if this correction is not applied, large discrepancies occur in the vertical direction. For example, if point cloud height data are compared directly with the elevation values on design drawings, discrepancies such as "tens of meters (tens of feet) higher (or lower) than expected" can arise.

As a countermeasure, always check and unify the coordinate systems of the datasets before performing alignment. If the coordinate system of the design data is unknown, examine the drawing legend and the notes on survey deliverables, and contact the designer or client for confirmation if necessary. Then convert the point cloud data to the appropriate coordinate system. If the point cloud was recorded in the instrument’s local coordinate system, perform a coordinate transformation (localization) using known control points on site to align it with a public or design coordinate system. Specifically, calculate offsets, rotation angles, and scale factors from the differences between point cloud coordinates and drawing coordinates at multiple known points, and apply a global correction such as a Helmert transformation. For elevation as well, use a geoid model to convert ellipsoidal heights to orthometric heights so that the vertical reference matches. Also be sure to check the unit system: some CAD data may have coordinates recorded in millimeter units (mm / in), and if compared directly with meter-based point clouds (m / ft) this will appear as a 1000-fold discrepancy. By resolving these differences in coordinate systems, datums, and unit systems in advance, you can correctly overlay the point cloud with other data.

Cause 3: Insufficient overlap or lack of feature points

Insufficient overlap (overlapping areas) between point clouds, or a lack of feature points that serve as cues for alignment, are also causes of misalignment. When stitching multiple scans together, if there is not a sufficient overlapping region between each scan dataset, the software cannot accurately compute their relative positions. For example, when scanning a wide flat terrain from two directions, if the overlapping portion is extremely small, there will be only a tiny common area to match, and even slight errors can make the alignment results unstable. Furthermore, if there is insufficient overlap between drone LiDAR flight paths, the resulting point clouds can exhibit striping offsets, causing adjacent datasets to misalign in height and position. Also, even when there is overlap, if most of the surface is bare ground and there are few distinctive objects or undulations to act as landmarks, the alignment algorithms may fail to converge properly, leaving local misalignments. Similarly, in environments such as building interiors where similar shapes repeat or in highly symmetric structures, automatic alignment can produce incorrect correspondences, resulting in parts being displaced.

As a countermeasure, from the planning stage of point cloud surveying, ensure sufficient overlap with adjacent scans. Generally, at least 20–30% overlap is considered desirable. If you feel the overlap is small, consider taking additional scans to cover the missing parts. In areas with few distinctive features, installing artificial targets (target boards or prisms, etc.) in appropriate positions can provide reliable correspondence points. When using targets, include at least three points in the field of view and arrange them spread out in the horizontal plane and also separated in elevation as much as possible, which helps suppress rotation and scale errors simultaneously. Avoid relying entirely on automatic registration; it is important to devise ways to obtain clear common references. Also, if the initial alignment (coarse registration) is significantly off, iterative algorithms such as ICP tend to converge to incorrect positions. As needed, assist the algorithm by manually aligning approximate positions or by specifying several correspondence points in the software so the algorithm can operate correctly.

Cause 4: Measurement Instrument Errors and the Effects of the Measurement Environment

The inherent accuracy limits of the measurement instruments used, as well as environmental factors at the site, can also affect the alignment accuracy of point clouds. In general, point clouds acquired with high-performance terrestrial laser scanners have errors on the order of a few millimeters to a few centimeters (a few mm / a few in), while point clouds from smartphone-integrated LiDAR can include errors on the order of several tens of centimeters (several tens of cm / several tens of in). In addition to such equipment-specific accuracy, if instruments are not properly calibrated or if sensors develop small shifts due to aging or temperature changes, the entire acquired point cloud may be slightly distorted. For example, when measuring targets tens of meters or more away (tens of m / tens of ft), laser pointing deviations and range correction errors can accumulate and appear as deviations of a few millimeters to a few centimeters (a few mm / a few in) at long distances. Also, for drone-mounted LiDAR, errors in position and attitude estimation from GNSS and IMU are reflected in the point cloud data. If RTK did not obtain a fixed solution during any segment of the flight, the point cloud for that portion will be locally displaced. Even for photogrammetry-derived point clouds, camera calibration errors or insufficient ground control points (GCPs) can cause scale errors or undulating deformations, compromising consistency with design data.

Furthermore, the measurement environment is also an important factor. When using satellite positioning, GNSS accuracy can be degraded by radio reflections (multipath) or reception blockage from surrounding buildings and trees, which can cause the entire point cloud to be offset. With laser scanners, measuring objects with extremely high or low reflectivity—such as mirrors or water surfaces—can produce noise points or ghost points that interfere with accurate alignment. If a tripod is shaken by strong winds or measurements are taken on scaffolding with significant vibration, parts of the point cloud can become blurred and shifted, potentially causing misalignment.

As a countermeasure, to prevent discrepancies caused by instrument accuracy and environmental conditions, first understand the characteristics of the equipment you will use and perform regular calibration and inspections. For measurements that require high precision, select measurement methods with the smallest possible errors (e.g., high-performance terrestrial LiDAR or drones equipped with RTK-GNSS). When using GNSS, improve positioning accuracy by setting up reference stations or using satellite augmentation signals, and take care to position with a FIX solution (integer fixed solution) whenever possible. Satellite geometry also affects accuracy, so choosing time periods when satellites are evenly distributed across the sky is a shortcut to improving accuracy and obtaining a FIX quickly. In environments where GNSS is unstable, such as confined sites or streets with high-rise buildings, switch to a terrestrial scanner or use solutions like a smartphone + GNSS device mentioned later to obtain stable coordinates. Also, before measurement, check the surrounding environment and, if there are unnecessary reflective objects or obstacles, remove them as much as possible or apply filters to remove noise points during measurement—thoroughly prepare in advance. During the post-processing stage of point clouds, remove clearly erroneous points (noise) before performing point cloud alignment so the algorithms are not misled by false detections. Eliminating each error factor arising from the measurement environment will lead to improved final point cloud alignment accuracy.

Cause 5: Mistakes in Data Processing or Configuration Errors

Human errors in the alignment process itself and software configuration mistakes are also causes that cannot be overlooked. Advanced point-cloud processing software provides comprehensive automatic registration features, but if those settings are incorrect or steps are skipped, the accuracy that should be achievable may not be obtained. For example, if you accidentally mix data from a different site when integrating different point clouds, or if you mistake which points should be matched, positional shifts will naturally result. There are also cases where applying a transformation while the coordinate system setting in the software is wrong causes unintended offsets. Furthermore, even if alignment succeeds for each pair of point clouds, small errors can accumulate during the process of chaining many scans, eventually producing overall distortion. This tends to happen when tiny misalignments at each step are left uncorrected as alignment proceeds to the next step, and because there are limits to applying a bulk correction (global adjustment) afterward, it is an error that should be detected and addressed early.

As a countermeasure, careful procedure management and verification are required during data processing. When handling multiple point clouds, organize metadata such as file names and coordinate systems to prevent confusion or mix-ups. Do not leave software alignment parameters (for example, ICP algorithm convergence criteria and tolerance settings) at their defaults; adjust them appropriately. After aligning each pair, always check the alignment accuracy. Calculate point-to-point distance errors in overlapping areas and differences between control points, and, if necessary, add corresponding points and readjust. If synthesizing 10 scans, it is important to finally verify that the first and last scans align properly and confirm that errors have not accumulated. Recent software includes global adjustment functions that optimally align all scans simultaneously. Use these functions as appropriate to minimize residual misalignment. In recent years, research using deep learning (deep learning) has advanced to achieve high-precision point cloud alignment under conditions that were previously difficult, and advanced software that can efficiently correct resulting misalignments has also appeared. If you have even the slightest doubt, do not compromise; the diligence of identifying and correcting causes directly leads to high-precision deliverables.

Summary

This concludes the explanation of the five main causes of point cloud alignment errors and the respective countermeasures. The handling of reference points, unifying coordinate systems, ensuring sufficient overlap, attention to equipment and environmental conditions, and precautions during data processing are all fundamental matters, but you can appreciate that neglecting even one can have a large impact on the results. In particular, on construction and civil engineering sites, errors of several centimeters (several inches) directly affect quality and safety, so be sure to apply the points raised here and carry out reliable point cloud processing. If you encounter alignment problems, reviewing these checkpoints in order should help you identify the cause and take appropriate measures. High-precision point cloud data will strongly support quality control and as-built verification on construction sites.

Even so, at some sites you may face issues such as "not being able to place enough reference points in confined spaces," "becoming too far from a base station for wide-area measurements," or "not being confident in coordinate transformations or software settings." In such cases, the iPhone-mounted high-precision GNSS positioning device "LRTK" can be helpful. LRTK consists of a compact, high-performance RTK-GNSS receiver and a smartphone app, and is a next-generation simplified surveying solution that allows anyone on site to easily achieve centimeter-level (half-inch-level) positioning. Even in environments where conventional methods tend to become unstable, stable high-precision positioning is possible through simultaneous use of multiple GNSS satellites and a built-in tilt correction function. Moreover, there are models that can directly receive signals from the Quasi-Zenith Satellite "Michibiki"'s centimeter-level (half-inch-level) augmentation service (CLAS), enabling standalone centimeter-level (half-inch-level) positioning via satellite correction information even at sites where a base station cannot be installed. It is also compatible with i-Construction promoted by the Ministry of Land, Infrastructure, Transport and Tourism, making it a tool that dramatically improves the productivity and accuracy of surveying work. Because you can check the deviation from reference points on the screen in real time while working, you can quickly notice and address coordinate system mismatches or positional shifts. If you are currently troubled by problems with point cloud alignment or complicated settings in your operations, why not consider using LRTK? It enables accurate alignment without advanced surveying knowledge and can become a reliable partner even under complex conditions.