How to create a deformation map from point clouds? Six checkpoints to improve accuracy

Creating a condition map is the starting point for assessing the condition of infrastructure facilities and civil engineering structures. If abnormalities such as cracks, spalling, material loss, settlement, joint openings, and deformations cannot be organized on drawings, it becomes difficult to share inspection results among stakeholders, and repair decisions and comparisons over time tend to become ambiguous. Traditionally, the common workflow was to make handwritten records on site while visually inspecting and then convert them into drawings while cross-referencing photographs, but in recent years there has been an increase in cases where point cloud data are used to create condition maps. Because the entire structure can be recorded as a surface, it is easier to reduce omissions in the record and to review details later, which is a major advantage.

However, simply having point clouds does not automatically produce a highly accurate deformation map. If the data density acquired in the field is insufficient, small deformations will be overlooked, and if coordinate registration is not consistent, comparisons across multiple time points or overlays with drawings will introduce errors. Furthermore, if visualization proceeds without adequately understanding noise and data gaps, there is a risk of depicting deformations that do not exist or, conversely, erasing important deformations. In other words, when creating deformation maps from point clouds, there are key points to check at each stage—measurement, data processing, interpretation, and visualization.

This article is aimed at practitioners searching for information on "deformation map point clouds." It organizes the key concepts to keep in mind when creating deformation maps from point clouds and explains six practical check items to improve accuracy. The content covers the entire workflow—from how to think about on-site acquisition conditions, to establishing rules for diagramming, to methods for organizing the results so they are finished in a state usable as deliverables. It is compiled around common on-site issues so that it is useful both for those considering adoption and for those already operating systems but experiencing variability in output quality.

Table of Contents

• First, outline the workflow for creating a deformation map from point clouds.

• Checklist item 1: Is the required density appropriate for the size of the anomaly you want to observe?

• Check item 2: Is there any discrepancy between the coordinate reference and the alignment

• Checklist item 3: Are you interpreting it with an awareness of blind spots and missing areas?

• Check item 4: Is the shape being damaged by noise processing?

• Checklist item 5: Are the criteria for determining abnormalities and the rules for diagramming standardized?

• Checklist item 6: Has it been organized with the operation of the deliverables in mind?

• How to Combine On-Site Inspections When Creating Damage Maps from Point Clouds

• Summary

First, outline the workflow for creating a deformation map from point clouds

First, it should be understood that creating deformation maps from point clouds is not a simple three-dimensional measurement task but an information-organization job of accurately recording the as‑built condition and translating it into a form that can be conveyed as drawings. On site, structures are measured to obtain point clouds, and, as needed, registration and coordinate assignment are performed; thereafter the deformed areas are identified and reflected in two-dimensional drawings and digitized graphical data. If even one condition in this sequence of steps is left ambiguous, the reliability of the final deliverable decreases.

For example, if sufficient density cannot be secured on site, the continuity of cracks and the shape of their ends become ambiguous, increasing the amount of estimation required during the drafting stage. Conversely, even if the acquired point cloud looks clean, if alignment is unstable and misregistration between surfaces remains, the contours of delamination and steps can be misinterpreted. Furthermore, if the criteria for extracting deterioration vary between personnel, the quality of drawings produced from the same point cloud will differ. In other words, when creating deterioration maps from point clouds, it is necessary to consider not only measurement accuracy but also interpretation accuracy and operational precision.

In practice, it is important to proceed on the assumption that point clouds are not a panacea. They are highly effective for capturing surface geometry, but information that does not involve changes in shape—such as dirt or color variations—can be difficult to distinguish from point clouds alone. Furthermore, the interpretation of fine cracks can vary greatly depending on acquisition conditions and the presence or absence of auxiliary information. Therefore, as a basic approach, the final condition map should be created with the point cloud at its core while combining it with field observation records, photographs, existing drawings, and comparison with results from previous years.

With this premise shared, below we will go through six checklist items to improve accuracy. None of them are special approaches, but they are points that tend to be easily overlooked on site amid the busyness. Establishing these checklist items as a shared understanding from the outset makes it easier to reduce rework during the data collection stage and backtracking during the diagramming stage.

Checklist Item 1: Do the required density and the size of the defect you want to see match?

The first thing to check is whether the density of the acquired point cloud is sufficient for the size of the defects you want to capture. This is the most basic aspect, yet in practice it is the source of the most problems. If the spacing between points in a point cloud is coarse, small defects cannot be represented as continuous features. If you want to map crack widths, chipped corners, surface delamination, fine steps, and so on, there is a limit to how much careful post-processing can achieve unless the required density was met at the time of acquisition.

What’s important here is not to assume that you should capture the entire object uniformly at a high density. The required density varies depending on the type of structure and the type of deterioration. This is because the unit you need to examine differs depending on whether you want to understand settlement or deflection occurring over a wide surface, or to check cracks or localized loss on a wall surface. For the former, continuity of the overall shape is important, whereas for the latter, faithful reproduction of local details is important. Before entering the site, it is essential to organize which deterioration you want to record and with what level of precision.

Also, acquisition distance and incidence angle greatly affect point density. Even with the same equipment and the same settings, the actual reproducibility differs between surfaces measured from a distance and those measured up close. Surfaces captured at angles close to perpendicular are easier to capture accurately, whereas surfaces that can only be seen at oblique angles tend to have points that stretch or drop out, and their contours often become blurred. On site, rather than simply increasing the number of measurements, you need to be mindful of which surface you are measuring, from what angle, and how close you get when capturing it.

Furthermore, when creating damage maps, it is essential to consider whether the data will hold up under local magnification. Even if it appears acceptable in an overall view, when zooming into a local area for mapping the points may be sparse, weakening the justification for drawing continuous lines. From a practitioner's perspective, it is important to assess density not by how the data looks immediately after acquisition but by assuming the stage of producing deliverables where lines and areas are drawn. In other words, you must confirm that the point cloud is not merely visible but actually drawable.

If you neglect this check, you may need to re-acquire only the missing parts later, which could require revisiting the site. At facilities where revisits are difficult, or at sites with strict traffic controls, access restrictions, or challenging scaffolding conditions, it is not an exaggeration to say that the initial acquisition plan largely determines the quality of the results. If you want to improve the accuracy of deformation maps, it is important to treat point cloud density not as a result but as a condition to be designed from the outset.

Check Item 2: Are there any discrepancies between the coordinate reference and alignment?

Another important point is the consistency between the coordinate reference and the alignment. When creating deformation drawings from point clouds, the work often requires not only reproducing individual shapes but also overlaying them with existing drawings, comparing multiple time points, and checking differences before and after repairs. Therefore, if it is not clear which coordinate reference the acquired point cloud is based on and what procedures were used to align it, the reliability of the drawings will be reduced.

On-site, data acquired from multiple directions are sometimes merged into a single point cloud. Even when the alignment appears natural, small local misalignments can remain in the details. Symptoms such as surfaces appearing doubled, corners becoming blurred, or edges that should be straight appearing slightly wavy can be mistaken for deformation. In particular, caution is required because such misalignments tend to influence mapping decisions around delamination edges, joints, and step areas.

Also, when comparing multiple time points, slight coordinate inconsistencies can cause major misunderstandings. When you overlay last year’s and this year’s point clouds to look for changes, alignment errors can appear as subsidence or deformation. When creating deformation maps from point clouds, you must distinguish between apparent differences and actual changes. To do this, it is important to establish a comparable positional relationship using reference points, known points, or a stable reference plane.

Furthermore, when mapping deformations onto existing plan and elevation drawings, it is necessary to reconcile them with the drawings' reference system. On site, it is not uncommon for drawings to be outdated and not fully match current conditions. In such cases, treating the drawings as absolute and forcing an alignment can impose unnatural adjustments on the point cloud, resulting in a distorted representation of deformation locations. Conversely, if the point cloud alone is treated as the truth and the drawings' alignment is ignored, comparing with previous years' results can become difficult. What is important is to organize the process while clearly indicating which reference to prioritize and where the sources of error lie.

In practice, checking alignment accuracy is often seen as the job of specialists only, but those who produce deformation maps should also have at least a basic understanding. Verify whether there is any misalignment at corners, edges, flat surfaces, and areas with known dimensions, and if you can judge yourself whether the deviation is at a level that will affect the mapping, it becomes easier to reduce incorrect mapping. Remember that the accuracy of a deformation map is supported not only by the reader’s experience but also by the integrity of the coordinates and alignment.

Check Item 3: Are you interpreting while accounting for blind spots and missing areas?

One thing that is easy to overlook when creating deformation maps from point clouds is convincing oneself that areas that were not captured have been captured. Point cloud data can at first glance appear to cover a wide area, but in reality there are always blind spots and gaps. In particular, the back sides of protruding parts, the deep parts of recesses, narrow sections, areas in the shadow of attached objects, and locations obscured by vegetation or temporary structures tend to have poor point coverage. Those responsible for creating deformation maps need to interpret the data not just by looking at the information shown in the point cloud, but after understanding which areas are visible and which are not.

This issue is important because missing points can appear as actual deformation. For example, if points are absent from a portion of a wall surface, it may look like delamination or material loss. However, when viewing only the point cloud, it can be difficult to determine whether that represents real damage or simply insufficient capture. Conversely, when there is genuine deformation, it can be overlooked as a capture failure if the surrounding point cloud is coarse and the outline is unclear. In short, missing data and deformation are easily confused, and if you proceed with visualization without sorting this out, the quality of the deliverables will be unstable.

An effective countermeasure is to anticipate in advance the parts that are likely to become blind spots during the acquisition stage and to have a policy for supplementary capture. Furthermore, at the visualization stage it is useful to check not only the point cloud but also site photographs, records of capture directions, and the history of acquisition positions. If you know from which position the data was captured, it becomes easier to determine why there are no points in that area. When creating damage maps, it is very important to adopt the stance of treating what is not visible as still not visible.

Also, it is important to decide as part of the diagramming rules how to handle missing elements. For example, if it is unclear whether unreadable sections should be treated separately, extracted for on-site reinspection, or included in the diagram when supplemental photos are available, different staff members will make different judgments. As a result, even at the same site this can cause variation between people in the density and scope of the damage diagrams.

In the field, it is easy to fall into the assumption that having a point cloud means everything is covered. However, in reality a point cloud is a record that is strongly influenced by the conditions under which it was acquired. Only by understanding the distribution of occlusions and missing data, and interpreting the data on that basis, can you better prevent overestimation and oversights. A highly accurate deformation map should be designed not only from the parts that were well observed, but also to include how to handle the parts that were not observed.

Checklist Item 4: Is the shape being preserved during noise processing?

To make point cloud data easy to visualize, removing unnecessary points and organizing the data are essential. However, while this preprocessing is useful, applying it incorrectly can erase the deformation itself or smooth its contours. To create a highly accurate deformation map, it is necessary to balance reducing noise with preserving traces of the deformation.

There are various types of noise. Typical examples include moving objects and workers, temporary structures, vegetation sway, water-surface reflections, shadow effects, and irregularities caused by surface materials. Leaving these as they are interferes with interpretation, but processing them mechanically under uniform conditions can sometimes remove even local bumps and chipped corners. In particular, discontinuities and localized changes are what matter in deformation maps. In other words, placing too much emphasis on general visual smoothness can cause the loss of information needed for deformation maps.

Edges of cracks and delamination, the rise of step edges, and missing edge sections inherently exhibit irregular, fine variations. These can easily appear as noise and, depending on processing conditions, are also parts that can be lost. Those responsible should consciously use both datasets adjusted for overall readability and datasets close to the raw data for detailed inspection. When setting the lines and ranges that define visualization criteria, it is desirable not to rely solely on processed data and to retain the ability to revert to pre‑processing information as needed.

Also, special care is required for processes involving surface generation and interpolation. Procedures that smoothly connect missing parts are effective for improving appearance, but they also carry the risk of creating surfaces that do not actually exist. A deformation map is, in a sense, a drawing that visualizes defects and abnormalities, so if drawings are made assuming conveniently filled-in surfaces, the true deformations will become ambiguous. It is important to choose data for visualization that prioritizes faithful reproduction of the current condition over visual attractiveness.

Furthermore, in practice it is important to keep a record of noise processing. If it is not clear under what conditions the noise was removed and to what extent it was processed, it becomes difficult to later explain the validity of the results. In inspection and maintenance work, there are situations where you must be accountable for explaining the results. Considering that even a single line on a damage map may require a rationale for judgment, noise processing should be treated not merely as a task but as part of quality control.

A point cloud that is visually tidy and well-organized is not necessarily the same as a point cloud that accurately represents deformations. Rather than deciding processing parameters solely for ease of visualization, it is essential for improving accuracy to proceed while confirming how much of the deformation’s contours and local shapes have been preserved.

Checklist Item 5: Are the criteria for judging abnormalities and the rules for diagramming standardized?

In work that creates deterioration maps from point clouds, standardizing the assessment criteria and the rules for graphical representation is as important as the technical processing. No matter how high-quality the point cloud data is, if what is treated as a deterioration and how it is depicted on drawings are not standardized, the reproducibility of the deliverables cannot be ensured. Differences in experience and judgment among personnel will be reflected directly in the drawings, making comparison and sharing difficult.

For example, should areas with differences in surface color be treated as pre-delamination groups, or should they not be mapped if there is no change in shape? Should small chips be drawn as independent deterioration features, or grouped as part of larger nearby deterioration? When the ends of cracks are ambiguous, should you draw only the visible portions, or extend them by cross-referencing photographs? If these judgments are not standardized, the representation density on drawings will be uneven even within the same facility.

A condition map is not merely a diagram of abnormalities but a fundamental reference for management decisions. Because it is used for assessing repair priorities, making comparisons over time, and passing on inspection histories, it is important that anyone can interpret it in a consistent way. To that end, it is necessary to organize the extraction targets by type of condition, decide in advance the extent of items to be mapped, determine which documents will be used as supporting evidence, and establish how to handle unreadable portions.

It is also important to clarify the unit of depiction. Whether a deformation is represented by a line, enclosed as an area, drawn as a centerline, or depicted down to the end shapes will change how the results are interpreted. Because point clouds contain a large amount of information, there is a tendency to want to add fine detail, but excessive representation for management purposes can actually make drawings harder to read. Conversely, oversimplifying reduces the value of using point clouds. A design philosophy is needed about what to detail and what to organize and show.

Furthermore, when multiple people are working, conducting trial mapping to check for differences is effective. If you each map the same area separately, you can see where judgments diverge. By reviewing the rules based on those differences, it becomes easier to reduce variation in the final deliverables. Creating deformation maps from point clouds may appear to be an individual skill, but in fact it also involves developing rules.

Improving precision is not simply about drawing things more finely. True quality improvement is creating a state in which outcomes do not vary significantly regardless of who produces them. Unifying the evaluation criteria and diagramming rules is a confirmation point you should always secure as the foundation that supports the reliability of deliverables.

Checklist Item 6: Have the deliverables been organized with how they will be operated in mind?

The final item to check is whether the created deformation maps and the original data have been organized so they can continue to be used in practice. In work that produces deformation maps from point clouds, the process often does not end with simply submitting drawings once; those outputs are frequently used afterward for maintenance management, reinspection, repair planning, and preparation of explanatory materials. If you compile deliverables without keeping this in mind, the high-quality data you have acquired will not be utilized in the next fiscal year.

A common issue is that the relationships among raw data, processed data, visualization outputs, site photos, and work notes are not organized. Even if everything is assembled at the time of deliverable submission, when you review it six months or a year later, reuse becomes difficult if it’s unclear which point cloud was used to produce which deformation map. In particular, for maintenance management tasks that rely on longitudinal comparisons, it is important to organize materials so that differences between measurement dates can be tracked.

Naming rules and storage units are also points to review. If you organize the target facility name, measurement date, coverage, surface orientation, processing stage, deliverable type, and so on according to a consistent rule, it will be easier to search later. On site, the ability to find files itself becomes a measure of quality as the number of files grows. Because point clouds tend to have large file sizes, it is important to structure storage so that required deliverables can be accessed immediately.

Also, at sites where distress maps are used, it is not always possible to handle three-dimensional data itself. It is necessary to present the information in a way that can be understood by people with different viewing environments, such as managers, clients, and construction personnel. Therefore, in addition to the distress map itself, making it clear which distress is at which location and how it corresponds to photos and drawings makes sharing easier. Deliverables that are strong in practical work are valued more for being usable without hesitation by the people who need them than for being highly advanced.

Furthermore, organizing with future reacquisition in mind is also important. To enable comparisons over the same area and by the same criteria at the next inspection, recording data acquisition conditions, the rationale behind the criteria, and points to note will contribute to stable quality across years. To prevent efforts to create deformation maps from point clouds from becoming a one-off task, it is necessary to treat the deliverables as operational assets.

A highly accurate deformation map cannot be evaluated solely by its level of completion at the moment it is created. Quality encompasses its reusability, comparability, and explainability. With this perspective, point cloud utilization shifts from merely improving drafting efficiency to building a foundation for maintenance and management.

How to Combine On-Site Inspections When Creating Deformation Maps from Point Clouds

Up to this point we’ve reviewed six checklist items, but what’s especially important in practice is not to rely solely on point clouds. Point clouds are a very powerful means of documentation, but they do not completely replace on-site verification. Rather, to make the most of point clouds it is important from the outset to consider together what to confirm on-site and where to make supplementary records.

For example, locations where shape changes are small but should be watched as signs of deterioration, locations where surface information is important such as color changes or water leakage stains, and locations where you want to judge structural significance in relation to interfaces with ancillary equipment are situations in which on-site inspection has high value. This is because, even if a point cloud can record a surface, there are cases in which it is difficult to determine the nature of an anomaly. Conversely, point clouds are highly effective for capturing continuous shapes over wide areas, for later reinspection, and for sharing among multiple people. In other words, on-site inspection and point clouds are not in competition but rather have complementary roles.

In practice, it is realistic to first record the entire site comprehensively with point clouds and to supplement important areas or locations where judgments may differ with on-site notes and photographs. At the diagramming stage, organizing the whole based on the point cloud while corroborating areas with weak grounds for judgment using auxiliary information makes it easier to balance accuracy and efficiency. This method reduces on-site time while improving verifiability in later stages.

Also, to establish on-site workflows for producing deformation maps from point clouds, it is important to stabilize how positional information is handled at the time of acquisition. If it is clear which coordinates the target area was recorded in and from which position it was inspected, reflecting that in the deformation map and tracking during revisits becomes easier. In situations where you want to streamline on-site position checks and simple surveying, systems that enable handling high-precision positional information using a smartphone can be a great help. If there is a system—like LRTK—that can be attached to an iPhone to achieve centimeter-level positioning (half-inch-level positioning), it becomes easier to verify control points, grasp on-site coordinates, and organize capture and inspection positions, making the operation of point clouds and deformation maps more practical to manage. Creating the deformation map itself is work of point cloud processing and diagramming, but by streamlining the on-site checks and position management that occur before and after that, the overall quality and reproducibility can be further improved.

Summary

When creating a deformation map from a point cloud, the important thing is not to be reassured merely by the fact that the point cloud was acquired. To improve accuracy, you need to check each item one by one: whether the density required for the deformation you want to observe has been secured; whether there are problems with the coordinate reference and alignment; whether you are interpreting the data with an understanding of blind spots and gaps; whether noise processing has not destroyed important shapes; whether the criteria for judging deformations and the rules for mapping are unified; and whether the deliverables have been organized in a form that can withstand future operation.

Deformation drawings are not merely drawings; they are records that form the basis for maintenance and management decisions. That is why, more than the visual appeal of point clouds, what matters is what can be explained, what can be compared, and what can be handed over. Point clouds are a powerful means of broadly preserving on-site information, but to translate their value into tangible outcomes it is essential to design the entire flow—from acquisition, through diagramming, to operation—as a unified process.

At sites that are preparing to scale up their use of point clouds, it is effective to review the entire operational workflow — rather than treating deformation-map creation as an isolated task — to include position verification, field recording, coordinate management, and reinspection. In particular, if you want to streamline on-site coordinate checks, control-point surveying, and sharing of inspection locations, adopting a high-precision positioning device that mounts on an iPhone, such as LRTK, can greatly simplify the practical work before and after point cloud acquisition. To stabilize the quality of deformation maps, it is important not only to have point-cloud processing expertise but also to establish a system that secures correct positions on site and enables necessary information to be collected without strain. If you want to achieve both labor savings and improved accuracy on site, it is worth considering the use of such high-precision positioning.