7 Tips for Creating Point Clouds Using Control Points｜What to Check to Prevent Misalignment

There are many sites that struggle with issues such as the position not matching the expected location despite using control points, local undulations, distortion only at the edges, or cross-sections appearing unnaturally high. It is often assumed that adding control points will automatically yield high accuracy, but in reality a stable point cloud is achieved only when the number of control points, their placement, observation accuracy, shooting conditions, and the way constraints are applied during processing all align. Conversely, even if the control points themselves are not wrong, incorrect placement or usage can lead to the troublesome situation where the overall alignment appears consistent while only parts are misaligned.

Especially in practical work, it is assumed that point clouds will be used in downstream processes such as as-built verification, current condition surveys, earthwork volume calculations, cross-section generation, and overlaying with design data. Therefore, it is not enough for them to merely look good; it is important that they can be trusted as coordinates. In this article, aimed at practitioners searching for "標定点 点群" ("control points point cloud"), we explain the basics for preventing misalignment when creating point clouds using control points, and we organize seven practical tips to keep in mind on site. This is not simply about increasing the number of points; from a practitioner’s perspective we summarize where to review to make accuracy more stable, including the order of checks.

Table of Contents

• Main reasons why point clouds are misaligned even when using control points

• Tip 1 First, clarify the purpose of the control point and don't confuse its roles

• Tip 2: Place control points evenly around the perimeter and at points where elevation changes

• Tip 3: Establish the coordinate accuracy of your control points before pursuing the accuracy required of the point cloud.

• Tip 4: Make reference points easy to capture and easy to read to ensure visibility

• Tip 5: Ensure overlapping shots and organized shooting routes so you don't rely solely on control points

• Tip 6: In analysis, apply constraints while balancing calibration points and camera parameters.

• Tip 7 Don't miss local misalignments by checking verification points and cross-sections

• Summary

Main reasons point clouds are misaligned even when using control points

When point clouds are misaligned despite using control points, many sites tend to assume that there aren’t enough control points. However, in reality, insufficient point numbers are not the only cause. More often, the problem is that the roles and placement of the control points do not match the purpose of the work. For example, even if you think you’ve spread them out in plan, if you haven’t adequately secured the perimeter of the survey area or placed reference points in locations with large elevation differences, you’re likely to see misalignment where only the interior aligns while the edges appear to float.

Another easily overlooked point is when the coordinates of the control points themselves are unstable. Point cloud processing is driven by the reference information you input. If you try to force a tight alignment while the control point coordinates remain unstable, you may end up deforming the overall shape and causing local distortions. Even if it appears that the point cloud has been fitted to the control points, the shape may be deviating from the actual object. Because this is difficult to judge by appearance alone, it is a troublesome problem.

Furthermore, if shooting conditions are poor and the correspondences between images are weak, the overall geometry becomes unstable even when control points are present. If image overlap is insufficient, many similar patterns make it difficult to extract feature points, or some images suffer reduced quality due to backlighting or shadows, the analysis ends up relying solely on the control points. In such a state, phenomena such as localized stretching, sagging, or twisting are more likely to occur.

In other words, control points are not foolproof. Control points provide absolute coordinates to a point cloud and are an important element in stabilizing the entire dataset, but they do not by themselves guarantee accuracy. Treat the planning, observation, image capture, analysis, and verification of control points as a single, continuous workflow, and separately consider where errors are likely to be introduced — that is the first step to preventing misalignment.

Tip 1 First, clarify the purpose of the control points and avoid confusing their roles

The first thing to be conscious of is to clarify why you are placing control points. In practice, although the term "control point" is widely used, points that provide coordinates, points used to verify accuracy, and points used to assist alignment can be mixed together in use. If work proceeds while these roles remain ambiguous, you may not have enough points in the places where they are needed and later be unable to perform accuracy checks.

There are two main purposes for using control points. One is to place the entire point cloud into a known coordinate system. The other is to independently verify whether the created point cloud is truly accurate. If you try to accomplish both with only the same points, they may appear to fit well in analysis but it becomes difficult to judge whether the required accuracy has actually been achieved. If you want a point cloud that can be used with confidence in practice, it is important to separate the points used for alignment from those used for verification.

Also, what should be emphasized changes depending on the object or deliverable. For as‑built surveys where planimetric position is important, and for volume calculations and cross‑section generation where vertical reproducibility is important, the placement of control points to be prioritized differs. Even if control points are aligned planimetrically, weak vertical performance will create inconsistencies in longitudinal profiles; conversely, if you focus only on vertical accuracy and leave planimetric bias unaddressed, offsets will become apparent in plan views and when overlaying designs. You should decide what to prioritize according to the purpose before planning the control‑point layout.

If the project team aligns its understanding at this stage, later processes will be more stable. If the field staff "placed targets where they were visible," the surveyors "measured only where they could," and the analysts "used the points included as-is," these local optimizations can accumulate and easily undermine overall optimization. To prevent such misalignment, it is important to share the view that control points are not merely on-site markers but design elements that support point cloud quality.

Tip 2: Arrange control points evenly around the perimeter and at points of elevation change

The most important thing when placing control points is to cover the target area as a surface. If control points are clustered near the center, that area may be stable, but the outer perimeter and edges are likely to exhibit shifts or distortions. Because a point cloud reconstructs the entire object in three dimensions, you need to properly constrain not only the center but also the corners and boundary areas. Especially for wide areas or elongated targets, if there are few reference points at the ends, errors tend to be amplified the farther away they are.

A common mistake here is concentrating control points only in locations that are easy to install. If points are biased toward places that are safe to place, easy to walk on, or easy to see, it may make fieldwork easier but will be disadvantageous for analysis. By placing points not only in well‑visible spots but also so as to capture the object's contours, it becomes easier to limit overall twisting and stretching. Often, it is more effective to first review whether there is any bias before increasing the number of points.

Even more important is the placement of control points in areas with elevation differences. On sites with changes in elevation—such as slopes, steps, cut-and-fill boundaries, and above and below structures—the three-dimensional shape becomes unstable if vertical constraint is weak, even if the points are well distributed in plan. By placing control points with attention to locations where elevation changes occur, it is easier to improve not only planimetric but also vertical reproducibility. You should avoid a situation where, despite large terrain changes, all control points are at similar heights.

A basic approach to placement is to secure the perimeter, add auxiliary points inside as needed, and also include representative points that capture elevation differences. Increasing the number of points itself is not the objective; it is important to be aware of which degrees of freedom are being constrained at which points. Since point cloud misalignment often results from biased placement, developing the habit of reviewing control point placement with both the plan and elevation drawings in mind before analysis will help stabilize accuracy.

Tip 3: Ensure the coordinate accuracy of control points is set before the accuracy required for the point cloud

No matter how you refine the processing parameters, if the coordinates of the reference control points are unstable, the overall quality of the point cloud will be limited. This is an extremely basic matter, but it is often put off when the site is busy. The coordinates assigned to the control points should be more reliable than the accuracy required of the point cloud results. If this is ambiguous, you cannot correctly judge the quality of the analysis results.

For example, if the object was not adequately secured during measurement, if it was strongly affected by the surrounding environment, or if the mounting surface itself was unstable, the coordinates of the control points will show variation. In that situation, if you force the fit to the control points during analysis, the residuals may appear small at first glance, but in reality the shape may be sacrificed. Rather than being reassured by only looking at the differences between the control points and the point cloud, you should first verify whether the reference itself is stable.

Also, the handling of coordinate systems and unit systems cannot be overlooked. If planar coordinates and vertical datums are mixed, or if on-site local coordinates and known coordinates are treated ambiguously, consistency can be lost after processing. Even if the values can be read in numerically, if the assumptions about the datums are off, the resulting point cloud will be difficult to use as a deliverable. Before entering the values of the control points, it is necessary to clarify which coordinate system, which vertical datum, and which units will be used.

In practice, workflows tend to prioritize efficiency by shooting first and then trying to align things later using control points. However, if you proceed with the quality of the control points left ambiguous, it becomes difficult to isolate the cause later. When discrepancies arise, it becomes hard to tell whether the issue is with the shooting, the analysis settings, or the control-point coordinates. To produce a stable point cloud, the quickest route is ultimately to reduce error factors as much as possible during the control-point observation stage and to secure the reliability of the reference beforehand.

Tip 4 Make control points easy to capture and read to ensure visibility

Calibration points require more than just correct coordinates; it is extremely important that they be clearly recognizable in the image. On site, even if the person who installed them can see them, during the image-processing stage the contours often become indistinct and hard to read. If they blend into the background, are obscured by shadows, or appear distorted by the viewing angle, errors can easily enter into the specification of the calibration point positions themselves.

Particular attention should be paid to the relationship between the size of control points and the flight altitude or shooting distance. Even if they appear sufficiently large on site, in images taken from overhead or from a distance they may occupy only a few pixels. In such cases, the error in reading the center position becomes large, which can result in unstable accuracy despite using control points. Control points should be made in sizes and shapes that can be read without difficulty given the target area and shooting conditions.

Also, the surrounding patterns and ground surface conditions cannot be ignored. On ground with many similar colors, vegetation, gravel, highly reflective surfaces, or dark pavements containing moisture, control points tend to become obscured. Control points with poor visibility are prone to inconsistent interpretation even when read manually, increasing the uncertainty when matching the same point across different images. If processing proceeds while the positions of control points remain ambiguous, the analytical constraints become weak and this leads to misalignment.

On site, practical measures include ensuring contrast with the background, choosing shapes whose centers are easy to identify even when viewed at an angle, and verifying legibility with test shots before photographing. Control points should be designed not only with placement in mind but also to ensure they are properly captured. Rather than struggling later to read poorly visible control points, creating ones that are easy to read from the start will greatly improve both accuracy and workflow efficiency.

Tip 5: Create overlap in imagery and organize shooting routes so you don't rely solely on control points

In point cloud generation, while control points are important, if the connections between images are weak the overall geometry will not be stable. A common misconception is that firmly placing control points can compensate for somewhat poor image capture quality, but in reality there are limits. Control points serve to provide absolute coordinates, but it is ultimately sufficient overlap and stable image capture that support the geometric relationships of the entire image set.

If capture overlap is insufficient, there will be fewer matching points between adjacent images and the local connectivity will weaken. As a result, even if areas near the control points are correct, sagging or shifts can easily occur in the sections between them. In particular, on monotonous ground surfaces, structures with many repetitive patterns, or areas with strong shadows, image connectivity becomes more unstable than expected. In such sites, rather than using a single, simple shooting path, combining crossing directions and viewpoints at different heights appropriately helps improve the stability of the reconstruction.

Sudden altitude changes and variations in capture intervals are also factors that cause point cloud misalignment. If close images and distant images are mixed too much, the appearance of feature points changes significantly, making correspondences prone to instability. Shooting at a steady pace, not missing parts of the subject, and ensuring sufficient overlap across the required area will, as a result, maximize the effectiveness of control points. There are cases where good capture can produce stable results even with few control points, but it's safer to assume the opposite—that many control points will not compensate for poor capture—is unlikely.

Furthermore, on sites with many moving elements during capture—such as wind, people moving, vehicles, or changes to temporary structures—some images may suffer reduced quality. When such images are mixed in, they can disrupt the overall consistency of the analysis and make alignment to control points unstable. It is important not only to increase the number of images but also to exclude unnecessary images and ensure consistent overall quality. To prevent misalignment, reviewing whether the image network is stable before increasing the number of control points is essential.

Tip 6: Apply constraints during analysis while monitoring the balance between calibration points and camera parameters

What you need to be careful about during the analysis stage is not to make aligning to the calibration points an end in itself. If you tighten constraints too much in order to reduce calibration-point residuals, the strain can be shifted onto other degrees of freedom. A typical example is when the internal degrees of freedom of the camera or the pose estimation are over-adjusted, resulting in an unnatural overall shape of the point cloud. If the calibration points match but you observe phenomena such as surfaces warping, edges floating, or cross-sections undulating, the balance may have been lost.

What is important is to comprehensively consider the errors of the control points, image consistency, and the naturalness of the shape. If you judge only by the residuals of the control points, it becomes difficult to notice an unreasonable solution. In particular, caution is required when the result is being strongly pulled by a small number of control points or when only some points are having an extreme influence. You must also check together whether the whole forms a natural three-dimensional shape, whether independent check points are not showing inconsistencies, and whether there is any oddness in cross-sections or plan views.

Deciding how much freedom to allow in camera settings is also important. The optimal approach differs for each site, but you cannot simply allow everything to be freely adjusted. If the degrees of freedom are too high, adjustments made to fit the control points will absorb elements that should not change, and as a result the reconstructed shape will be distorted. Conversely, if you fix parameters too rigidly, you may not be able to accommodate real-world differences, and alignment may fail. What is needed is not extreme fixation or extreme freedom, but a balance appropriate to the object conditions and the shooting conditions.

Therefore, in analysis you should avoid making a one-shot decision and instead adopt an attitude of adjusting while watching the selection and usage of control points, the distribution of residuals, the appearance of the point cloud, and any anomalies in cross-sections. Even if the numbers look good, you should not take the results at face value if there are local distortions. Conversely, even if control point residuals appear somewhat large, if the overall shape is stable and the check points also yield reasonable results, that outcome may be more reliable in practice. It is important to understand that preventing discrepancies is not about chasing a single metric, but about maintaining consistency across the entire analysis.

Tip 7: Don't overlook local misalignments by checking verification points and cross-sections

Misalignments in point clouds can be difficult to notice when viewing the whole. Even if it appears natural from above, shows no oddities in a colored display, and the control-point residuals are not that bad, slicing cross-sections can reveal unnatural waviness. Therefore, in the final check it is important to have validation points separate from the control points and to verify not only planar alignment but also cross-sectional and vertical consistency.

The role of validation points is to verify accuracy against independent references that were not used in the analysis. If you evaluate only with the points used in the analysis, it will of course appear to agree. To know the actual quality, you need to check with points that were not involved in the processing. This makes it easier to detect excessive fitting and local distortions. In particular, it is effective to place validation points at locations prone to misalignment, such as the outer perimeter, edges, areas with elevation changes, and points where structures transition.

Also, cross-section checks are extremely useful in practical work. Vertical distortions that are hard to see in plan view, surface warping, and unnatural sagging of slopes become easier to grasp when viewed in cross-section. Rather than finishing with simply looking at the completed point cloud, checking it while keeping in mind the cross-section orientations and critical locations that are intended to be used as deliverables will help prevent rework in later stages. When the purpose is clear—such as creating as-built drawings, overlaying designs, or calculating quantities—you should evaluate it from a perspective close to that intended use.

Also, when a misalignment is found, it's important to have a perspective for isolating its cause. If only the perimeter is misaligned, it's a placement issue; if only the elevation is unstable, it's insufficient constraint in areas of height difference; if there is local distortion, it's likely an issue with capture quality or the degrees of freedom in the analysis. Working backward from the symptoms and reviewing the setup accordingly makes it easier to achieve improvements. Point cloud generation does not always come out perfect on the first try. Precisely for that reason, carefully carrying out the final validation step and ensuring you are in a position to judge which conditions need to be modified is indispensable for stable operation.

Summary

To prevent misalignment when creating point clouds using control points, you need to consider not just placing the control points but the entire workflow of how you design them, observe them, analyze them, and validate them. If you place points without clarifying their roles, you will lack references where they are needed. If placement is biased, distortions tend to appear at the outer edges and in areas with elevation differences. If coordinate accuracy is unstable, the very frame of evaluation for the analysis results will wobble. Poor visibility increases the chance of specification errors on the images. If capture quality is insufficient, control points alone cannot support the overall shape. If you over-constrain during processing, the numbers may fit but the shape can be distorted. And if you omit verification points or cross-section checks, you risk overlooking local misalignments.

In other words, stabilizing point cloud accuracy through control points is not only a matter of the number of points, but also a matter of the quality of planning and operations. To produce reproducible results in the field, it is important to standardize the design before placing control points, the management during capture, and the evaluation after analysis, and to verify them from the same perspectives each time. In practical work especially, what is required is to create point clouds that can be relied on in downstream processes, rather than point clouds that simply look clean. If you have checklist items in advance to prevent misalignments, you can reduce rework and re-surveys and more easily improve the efficiency of the entire operation.

In sites where you want to carry out control point installation and coordinate verification more reliably, the ease of on-site position checks and simple surveying can make a big difference. In such situations, using LRTK, an iPhone-mounted GNSS high-precision positioning device, can help streamline tasks such as verifying control point installation locations, determining on-site coordinates, and recording surrounding points. As a preparatory process that supports the quality of point cloud generation itself, if you want to improve the accuracy and operability of work around control points, introducing an easy-to-use high-precision positioning system like LRTK on site leads to more consistent results.