Improving AR Integration Accuracy with LRTK High-Precision Point Cloud Acquisition and Gaussian Smoothing

Background: Challenges and Opportunities in AR and Point Cloud Integration

Augmented reality (AR) technology is attracting attention in construction and surveying fields. AR, which overlays virtual 3D models and information onto a site simply by pointing a smartphone or tablet, is a powerful tool for intuitively sharing design drawings and CIM models. However, a major challenge for practical AR use is the accuracy of alignment with the real environment. For example, when projecting a building’s design model onto the actual site, even a shift of a few tens of centimeters between the model and the real structure can make the model appear to float or sink into the ground for users. Even slight display jitter can interfere with precise verification and work.

To solve this alignment accuracy problem, it is important to measure the real world with high precision and use it as the AR reference. One promising approach is the use of 3D point clouds. A point cloud is three-dimensional data composed of countless points acquired by laser scanners or LiDAR, which records real-world geometry in detail. Using a point cloud as the foundation for AR can potentially enable comparison with virtual models and detection of differences from the real shape at millimeter-level precision. Traditionally, however, acquiring point clouds required expensive equipment and specialized skills, and the captured point cloud data often contained noise and errors. These issues sometimes caused jitter and misalignment when combining point clouds with AR, preventing the expected accuracy improvements.

A recent solution to these challenges is high-precision point cloud acquisition via LRTK and the application of Gaussian smoothing (Gaussian filter) to that data. LRTK is a system that makes it easy for anyone to obtain high-precision point clouds with absolute coordinates using RTK-GNSS technology. Applying Gaussian-kernel-based smoothing to these point clouds can dramatically improve AR integration accuracy. This article explains the mechanism and effects in detail.

What Are RTK-GNSS-Referenced High-Precision Point Clouds Acquired with LRTK?

LRTK combines a smartphone’s (e.g., iPhone) built-in LiDAR sensor or camera with high-precision positioning using real-time kinematic GPS (RTK-GNSS) to create a point cloud measurement system. By attaching a lightweight dedicated receiver (about 165 g) to the phone and using a proprietary app, anyone can easily achieve centimeter-class positioning and 3D scanning that previously required specialized equipment. By using RTK-GNSS correction information, positional errors of about 5–10 m typical of normal smartphone GPS can be reduced to approximately ±1–2 cm horizontally and around ±3 cm vertically. Each point in the captured point cloud is assigned absolute coordinates consisting of latitude, longitude, and elevation, so the point cloud itself becomes a high-precision 3D map aligned to the site coordinate system (public coordinates or any local coordinate system).

LRTK allows such high-precision point clouds to be captured with remarkably simple steps. For example, even across wide roads or slopes, you can scan point clouds covering several tens to hundreds of meters in a short time (in some cases just 1–2 minutes) simply by walking while holding a smartphone. Because all captured point clouds contain global position information, no post-processing alignment is required. In other words, it’s like obtaining a “digital replica” of the entire site measured in situ. These LRTK high-precision point clouds are extremely effective as a basis for AR integration. If 3D design models (BIM/CIM, etc.) are created in the same coordinate system, simply overlaying the captured point cloud and the design model will result in a near-perfect positional match, eliminating cumbersome alignment tasks. Moreover, the ensured vertical accuracy reduces the likelihood of models appearing to sink into or float above the ground.

Spatial Noise in Point Clouds and Problems for AR Use

Even though point clouds acquired with LRTK are high-precision, sensor measurements inherently contain some noise and errors. Spatial noise in a point cloud refers to point dispersion that causes what should be flat surfaces to appear bumpy or concentrated shapes to be slightly scattered. Causes include LiDAR ranging errors, surrounding environmental factors (varying reflectivity or occlusion), and slight device motion. For example, a point cloud scanned from a perfectly flat concrete floor may still show height variations on the order of centimeters, spreading like a cloud. Using such noisy point clouds directly in AR can cause several problems.

First, when detecting or recognizing planes and objects in the real world based on point clouds, noise can lead to false detections and instability. AR apps detect planes in the environment to place virtual objects, but if the floor point cloud is uneven, it won’t be recognized as an accurate plane and virtual objects may wobble.

Second, when a point cloud itself is used as the background “reality” for AR rendering, point dispersion due to noise appears as AR overlay jitter. Imagine moving the device slightly and seeing the virtual model and point cloud shift subtly—this happens because local point errors cause the relative relationship between the virtual model and the camera viewpoint to vary irregularly.

Third, when comparing captured point clouds with design models to assess conformity, noise adversely affects the results. Slight discrepancies caused by measurement noise can appear even where things should match perfectly, making it difficult to determine what constitutes construction deviation versus measurement noise.

As described above, noise and dispersion in point clouds reduce AR integration accuracy. Even with high-precision LRTK point clouds, millimeter- to centimeter-scale variations cannot be ignored—on-site differences of several centimeters become clearly visible as misalignment or jitter in AR. Therefore, filter processing of point cloud data is necessary. Among these, smoothing—which averages each point’s position with its neighbors to produce a smoother result—is a fundamental and effective noise-reduction method. The next section explains the principle of Gaussian smoothing, a representative smoothing technique.

Basics of Gaussian Smoothing: Spatial Weights and Kernel Mechanism

Gaussian smoothing corrects each point’s values (such as position coordinates) on a point cloud by replacing them with a weighted average of neighboring points. The key is the weighting by a Gaussian kernel. The Gaussian kernel is a weighting function based on the familiar normal distribution (the “bell curve”) from statistics, where nearer points receive larger weights and farther points receive smaller weights. Specifically, a neighbor point at distance d from the target point is weighted proportionally to exp(-d^2/(2σ^2)) (σ is the standard deviation of the Gaussian distribution). By applying such continuous distance-dependent weighting, spatially smooth smoothing is achieved.

An intuitive image is placing each point on a rubber sheet and letting it be pulled by its neighbors so positions adjust relative to one another. Extreme isolated points (outliers) are hardly pulled by neighbors and thus their conspicuous noise is reduced. Dense clusters of points adjust mutually, producing smoother surfaces and curves overall. The advantage of Gaussian smoothing over a simple mean filter is that it allows continuous control of smoothing based on distance. Close points are mostly preserved while distant points have reduced influence, so edges can be preserved as much as possible while reducing noise. However, very sharp corners and fine details may become somewhat rounded; for AR integration purposes, the benefit of noise reduction typically outweighs the slight shape changes caused by smoothing.

Implementation and Parameter Tuning: Effects of Standard Deviation and Neighbor Selection

Applying Gaussian smoothing to point cloud data in practice requires several parameters and implementation considerations. The main points are the Gaussian kernel’s standard deviation σ (smoothing strength) and the method for selecting neighboring points used for smoothing.

First, σ directly controls the smoothing range and degree. A smaller σ causes the Gaussian weight to decay rapidly and averages only very close points, which removes fine noise while preserving small bumps and edges. Conversely, a larger σ assigns significant weight to more distant points, producing broader averaging. This smooths larger irregularities but may blur fine structural details like sharp edges. The appropriate σ depends on the noise level of the point cloud and the scale of the features of interest. For typical LiDAR point-cloud noise reduction, σ is often set on the order of millimeters to centimeters and adjusted while observing the effect. The important balance is to remove noise while preserving critical shape features—choosing neither too large nor too small a σ.

Next, neighbor selection determines which points around the target are considered its neighbors for smoothing. Typical approaches are using a fixed radius around the point or selecting the k nearest points. When using a radius, set it to be roughly similar to or larger than σ to include enough neighbors. If the radius is too small, insufficient neighbors yield unstable smoothing; if too large, unrelated points from different objects may be included and distort the shape. When choosing k nearest neighbors, k should be reasonably large (for example, 20 or 50 points) to obtain a stable average. However, when point density varies across the cloud, a fixed k can cause over-smoothing or under-smoothing in different regions, so combining radius and k constraints can help.

Computational efficiency is another practical concern. Performing neighbor searches and weight calculations for every point can be time-consuming for large point clouds. Using spatial indexes such as KD-trees for neighbor searches enables efficient computation. In LRTK system cloud processing, uploaded point clouds can be automatically denoised and smoothed so users receive consistent results without worrying about parameters. If processing locally, it is safe to start with relatively weak smoothing and gradually increase it as needed.

Examples of Reduced Jitter in AR When Using Smoothed Point Cloud Models

So what effect can Gaussian smoothing applied to point clouds actually have on AR rendering? Consider the example of a paved road’s point cloud. The original point cloud acquired by LRTK contained height variations of several centimeters even on a visually flat road surface. When displaying a virtual design line on the road in AR using that raw point cloud, slight vertical jitter of the virtual line was observed as the device moved. The virtual line was dragged by small surface irregularities in the road point cloud, causing it to appear to wobble with changes in viewpoint.

Applying a Gaussian smoothing filter to this road point cloud averaged out the small irregularities. Specifically, smoothing used a kernel with σ around 5 cm over a radius of several tens of centimeters. After smoothing, the road surface appeared as an almost ideal gentle plane, and the original centimeter-scale bumps were no longer prominent. Using this smoothed point cloud as the AR base, the previously wobbling virtual line became perfectly stable; even when moving the device, users did not perceive any jitter. Users reported that the virtual line looked as if it had actually been painted on the ground, demonstrating a natural overlay of reality and virtual content.

This case shows that Gaussian smoothing removes random error components from point cloud data and reduces AR overlay jitter. With LRTK’s globally referenced point clouds, positional offsets are minimal, so the remaining issue is fine noise. Averaging out that noise via smoothing results in a consistent appearance of the point cloud across consecutive frames, which in turn stabilizes the appearance of virtual objects. For on-site AR use, preventing display wobble as users walk or change gaze is important for safety and reliability; Gaussian smoothing contributes significantly to this practical quality improvement.

Application to Conformity Assessment Between Point Clouds and CIM/3D Design Models

Gaussian smoothing is effective not only for stabilizing AR displays but also for assessing conformity between point cloud data and design models (BIM/CIM). On construction sites, scanned point clouds of completed structures or terrain are compared with original design models to verify how closely the work matches the plan. For smooth assessment, it is desirable that the point cloud lacks excess noise so that only true differences against the design model stand out.

Raw point clouds without smoothing typically contain the aforementioned centimeter-scale variations. When overlaid with the design model, these noise-induced variations cause colored differences (e.g., non-zero errors on a distance map) to appear in areas that should match, making it harder to distinguish true construction deviations. In contrast, applying Gaussian smoothing reduces random variations substantially, making systematic differences clear on difference maps. For example, if a region of the point cloud is consistently 3 cm higher than the model, it becomes easier to interpret that as an actual 3 cm construction overbuild. Meanwhile, sub-centimeter differences that were obscured by ±3 cm noise in the raw cloud will no longer be displayed and can be regarded as negligible, reducing noise-related uncertainty in evaluations.

In addition, Gaussian kernels help with point cloud-to-model alignment (pose alignment). Common algorithms like ICP (Iterative Closest Point) compute a transform that minimizes distances between corresponding points, but noisy point clouds can lead to incorrect correspondences and unstable convergence. With smoothed point clouds, outlier points have less influence due to averaging, and global shape-based alignment becomes easier. In other words, smoothing local irregularities with a Gaussian kernel facilitates achieving global alignment.

Thus, Gaussian smoothing aids the entire process of comparing point clouds with CIM/3D models, enabling more reliable verification. When overlaying scanned site data with design files to decide, for example, “this area is slightly overbuilt compared to the drawing” or “this slope is gentler than planned,” smoothing reduces the risk of misjudgment due to noise.

Benefits of Introducing Gaussian Smoothing: Stable AR and High-Precision Matching

As discussed, combining high-precision point clouds with Gaussian smoothing brings a qualitative leap to AR and point cloud integration. The introduction benefits can be summarized as follows:

• Stabilized AR overlays: Noise-induced jitter from point clouds is suppressed, so virtual models appear fixed to the real environment. There is less need for markers or manual adjustments for alignment, allowing users to use AR without stress.

• High-precision model matching: Smoothed point clouds with global coordinates enable high-accuracy overlap between site data and design models. As a result, differences between the model and reality can be measured accurately to millimeter- to centimeter-level, making AR suitable for quality control and as-built inspection.

• Improved consistency of scan data: Gaussian smoothing reduces inconsistencies and positional fluctuations between consecutively acquired point cloud frames. Large scanned datasets can be handled as a single smooth model, simplifying post-processing and analysis.

• Increased on-site work efficiency: With point clouds that require no alignment and stable AR, the ability to check measurements and give instructions on the spot is greatly enhanced. For example, showing a design model in AR during construction reduces surveying and inspection work and helps detect errors early.

Smoothed point cloud data thus significantly enhance AR reliability and precision, strongly supporting the bridge between digital and real worlds.

Future of On-site AR Using LRTK × Gaussian Smoothing

The combination of LRTK’s high-precision point cloud acquisition and Gaussian smoothing will likely drive AR use on site to the next level. Future developments may include:

• Real-time feedback: LRTK can provide centimeter-level positioning in real time. If fast point cloud smoothing is combined with this, it may become possible to construct a smoothed point cloud model on the fly while holding a device and immediately reflect it in AR. If surveying and verification become integrated in real time, instant checks and feedback can be applied throughout construction processes.

• Integration with AR glasses: While smartphones and tablets are the primary AR platforms today, widespread adoption of AR glasses could allow LRTK positioning data and smoothed point clouds to be streamed to glasses so workers can continuously view high-precision AR hands-free. For example, a BIM model could be overlaid stably through the glasses, enabling digital support for tasks that previously relied on experienced judgment and visual inspection.

• Cloud sharing and remote support: Sharing LRTK-captured point clouds and AR information via the cloud would let off-site managers and designers check model conformity on the site’s point cloud model and send instructions. Smoothed point clouds can also reduce data size and network load, making remote AR support more feasible.

• New quality control standards: If high-precision point clouds and AR become standard, inspection and control protocols may evolve. For example, alignment tasks previously done with reference strings or chalk lines could be performed using AR-projected guidelines, and as-built inspections might be judged via AR color-coded difference displays. Stable smoothed point clouds would be an essential component of such digital-first practices.

In these ways, LRTK’s high-precision positioning and Gaussian smoothing can greatly expand the practical uses of AR on site. AR with assured accuracy and stability may move beyond mere visualization and become an infrastructure for on-site work.

First Steps Toward Practical Use: Start Acquiring and Utilizing High-Precision Point Clouds with LRTK

We have described the effects of combining advanced point cloud processing and AR technologies. To realize these benefits in actual projects, obtaining high-precision point cloud data is the essential first step. With LRTK, absolute-coordinate point cloud measurement that previously required professional surveying equipment can be started easily with just a smartphone. Preparation and operation are simple, and untrained technicians can perform on-site 3D scans in a short time. Captured point cloud data can be automatically uploaded to the cloud and used on the spot for distance and area measurements and model overlays. If needed, Gaussian smoothing can be applied to reduce noise and further improve accuracy.

The combination of high-precision point clouds and AR is expected to become increasingly important in practice. As a technology that bridges the digital and real worlds, it has wide application potential in construction management, infrastructure maintenance, urban planning, and disaster prevention. As a first step, consider starting with simple surveying using LRTK. By visualizing site conditions precisely and intelligently smoothing and utilizing that data, you can achieve a stable AR experience and improved accuracy control.