High-precision Point Cloud Acquisition with LRTK and Improved AR Accuracy via Gaussian Smoothing

Background: Challenges and Potential of AR and Point Cloud Integration

Augmented reality (AR) technology is attracting attention in construction and surveying sites. AR, which allows virtual 3D models and information to be overlaid on the real site simply by pointing a smartphone or tablet, is a powerful tool for intuitively sharing blueprints and CIM models. However, a major challenge for practical AR use is the precision of alignment with the real world. For example, when projecting a building design model on site, if the model’s position differs from the actual structure by only several tens of centimeters, users may perceive the model as floating or buried in the ground. Even slight display wobble (jitter) can hinder accurate matching and work.

To solve this alignment accuracy problem, it is important to measure the real space with high precision and use it as the AR reference. This is where the use of 3D point clouds draws attention. A point cloud is three-dimensional data composed of countless points obtained by laser scanners or LiDAR, recording the real shape in detail. By using a point cloud as the foundation for AR, it becomes possible to detect differences between the virtual model and reality at millimeter resolution. However, point cloud acquisition has traditionally required expensive equipment and specialized skills, and the acquired point cloud data often contained noise and errors. These issues could cause wobble and positional discrepancies even when combining point clouds with AR, so the expected precision improvement was not always achieved.

A recent solution to these issues is high-precision point cloud acquisition using LRTK, combined with applying 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. By applying Gaussian-kernel smoothing to these point clouds, the accuracy of AR integration can be dramatically improved. This article explains the mechanisms and effects in detail.

What Are RTK-GNSS-referenced High-precision Point Clouds Acquired by LRTK?

LRTK is a point cloud measurement system that combines a smartphone’s (for example, an iPhone’s) built-in LiDAR sensor or camera with high-precision positioning using real-time kinematic GNSS (RTK-GNSS). By attaching a dedicated receiver weighing about 165g to the smartphone and using a proprietary app, centimeter-level positioning and 3D scanning that once required specialized equipment can be achieved easily by anyone. Leveraging RTK-GNSS correction information reduces the positional error that a normal smartphone GPS would typically have—on the order of 5–10 m (16.4–32.8 ft)—to about ±1–2 cm (±0.4–0.8 in) horizontally and about ±3 cm (±1.2 in) vertically. Each point in the acquired point cloud is assigned absolute coordinates consisting of latitude/longitude and elevation, making the point cloud itself a high-precision 3D map that aligns with the site coordinate system (public coordinates or any local coordinate system).

LRTK can acquire such high-precision point clouds with surprisingly simple procedures. For example, even for wide roads or slopes, you can scan point clouds over tens to hundreds of meters (tens to hundreds of ft) in a short time—sometimes in about 1–2 minutes—simply by walking while holding a smartphone. Because all acquired point clouds include global position information, no post-processing registration is required. In other words, it is like obtaining a “digital replica” of the entire site. These LRTK high-precision point clouds are extremely effective as a foundation for AR integration. If 3D design models (BIM/CIM, etc.) are created in the same coordinate system, simply overlaying the acquired point cloud and the design model results in a precise match, eliminating complicated alignment work. Moreover, guaranteed vertical accuracy reduces the likelihood of the model sinking into or floating above the ground.

Spatial Noise in Point Clouds and Issues for AR Use

Even though point clouds acquired by LRTK are high-precision, they are still sensor measurements and not entirely free of noise or errors. Spatial noise in a point cloud refers to point variations such as bumps appearing on surfaces that should be flat, or shapes that should be concentrated at a point being slightly scattered. Causes include LiDAR ranging errors, surrounding environmental factors (variations in reflections or occlusion), and slight device shake. For example, a point cloud scanned from a perfectly flat concrete floor can still show point heights fluctuating by several centimeters, spreading like a cloud. Using such noisy point clouds directly in AR causes problems.

First, when detecting or recognizing planes or objects in the real world based on a point cloud, noise can cause false detections and instability. AR apps detect planes in the environment to place virtual objects, but if the floor point cloud is bumpy, it will not be recognized as an accurate plane, causing virtual objects to wobble.

Second, when using the point cloud itself as the background “reality” for AR display, point scatter due to noise appears as AR overlay jitter. Imagine a situation where, by slightly moving the device, the virtual model and the point cloud appear to shift relative to each other. This is caused by local errors on the point cloud, making the relative position of the virtual model change irregularly with the camera viewpoint.

Third, when comparing the acquired point cloud with a design model to evaluate consistency, noise has a detrimental effect. Slight deviations caused by measurement noise can appear even where things should perfectly match, making it difficult to determine where construction error ends and measurement noise begins.

As described above, noise and point scatter in point clouds reduce accuracy when integrating with AR. Even with high-precision LRTK point clouds, it is not acceptable to ignore variations on the order of millimeters to several centimeters. In the real world, several centimeters are conspicuous in AR as clear offsets or wobble of objects. What is needed, therefore, is filter processing for point cloud data. Among these, smoothing—which averages each point’s position with its surroundings to make it smoother—is a fundamental and effective method for noise reduction. The next section explains the principle of Gaussian smoothing, a representative smoothing method.

Basics of Gaussian Smoothing: Spatial Weights and Kernel Mechanism

Gaussian smoothing is a process that smooths and corrects values (such as position coordinates) of each point in a point cloud by weighted averaging with neighboring points. The key is weight assignment using the Gaussian kernel. The Gaussian kernel is a weighting function based on the normal distribution (the familiar “bell curve”), giving larger weights to closer points and smaller weights to distant points. Specifically, for a neighbor point at distance d from the target point, a weight proportional to exp(-d^2/(2σ^2)) is applied (σ is the standard deviation of the Gaussian distribution). This continuous distance-based weighting enables spatially smooth smoothing.

An analogy is placing each point on a rubber sheet and adjusting positions so they pull on each other with the surrounding points. Isolated outlier points are hardly pulled by surroundings and thus their conspicuous noise is softened. Dense clusters of points correct each other’s positions, resulting in smooth surfaces or curved shapes. A strength of Gaussian smoothing is that, unlike a simple average filter, it allows adjusting the degree of smoothness according to distance. Close points are mostly preserved while the influence of distant points is suppressed, so edges and boundaries are preserved as much as possible while reducing noise. However, very sharp corners and fine details may become slightly rounded; for AR integration, the benefit of noise reduction often outweighs minor smoothing-induced shape changes.

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

When actually applying Gaussian smoothing to point cloud data, several parameters and implementation considerations are required. The main points are the Gaussian kernel’s standard deviation σ (the smoothing strength) and the method for selecting neighboring points used for smoothing.

First, the standard deviation σ directly controls the range and degree of smoothing. A smaller σ makes the Gaussian weight decay rapidly, averaging only very nearby points. As a result, fine noise is removed while small bumps and edges are relatively preserved. Conversely, a larger σ assigns significant weight to more distant points, producing averaging over a wide area. This smooths larger bumps but can blur fine details (such as edges on angular structures). The appropriate σ depends on the point cloud’s noise level and the scale of features of interest. For typical LiDAR point cloud fine-noise removal, σ is set on the order of several millimeters to several centimeters and adjusted while observing the effect. The important balance is to remove noise while preserving important features—avoid values that are too large or too small.

Next, neighbor selection determines which surrounding points are considered “neighbors” for smoothing each target point. Typical approaches use points within a fixed radius or the k nearest neighbors. When selecting by radius, the radius should be comparable to or larger than σ to include a sufficient number of neighbors. If the radius is too small, few neighbors are found and smoothing becomes unstable. If too large, points from unrelated objects may be included, distorting shapes. When selecting k nearest neighbors, k should be more than just a few points—setting it to a moderate number (for example, 20 or 50) yields a stable average. However, if point density varies across locations, a fixed k may over-smooth or under-smooth; combining radius and k constraints can be effective.

Implementation should also consider computational efficiency. Searching neighbors and computing weights for every point can be time-consuming for large point clouds. Using spatial indexes such as KD-trees for neighbor searches makes computation efficient. In LRTK systems, cloud processing can automatically perform noise removal and smoothing on uploaded point clouds, allowing users to get consistent effects without adjusting parameters. If processing locally, it is safer to start with relatively weak smoothing and gradually increase strength as needed.

Case Study: Reduced AR Jitter When Using Smoothed Point Cloud Models

So how effective is Gaussian smoothing in AR display? Consider a point cloud of a paved road. The original LRTK-acquired point cloud showed height variations of several centimeters even on visually flat pavement. When overlaying a virtual design line (for example, a centerline or finished elevation guide) directly on that point cloud in AR, the virtual line was observed to wobble slightly up and down as the device moved. The virtual line was dragged by microscopic bumps in the road point cloud, causing perceived instability with changes in viewpoint.

We applied a Gaussian smoothing filter to this road point cloud to average out fine bumps. Specifically, smoothing used a kernel with σ around 5 cm (2.0 in) over a radius on the order of several tens of cm. After smoothing, the road surface became an almost ideal smooth plane, and the original centimeter-scale bumps became unnoticeable. Using this smoothed point cloud as the AR reference, the previously wobbling virtual line became perfectly stable—users no longer perceived any jitter when moving the device. Users described it as “the virtual line looks as if it were actually painted on the ground,” indicating a natural overlap of reality and virtual content.

This example shows that Gaussian smoothing removes random error components from point cloud data, reducing AR overlay jitter. For LRTK point clouds with global coordinates, positional offsets are minimal, so remaining issues are fine-scale noise. Averaging out that noise via smoothing stabilizes the point cloud’s appearance across consecutive frames, and consequently the appearance of virtual objects. In on-site AR use, having displays that do not shake as users walk or shift gaze is important for safety and reliability. Gaussian smoothing makes a substantial contribution to such practical quality improvements.

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

Gaussian smoothing is effective not only for stabilizing AR display but also for evaluating consistency 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 reality matches plans. For this process to proceed smoothly, point cloud data should be free of excess noise so that only genuine differences from the design model are highlighted.

Raw, unsmoothed point clouds often contain several-centimeter variations across the site. When overlaid on a design model, noise can produce scattered colored differences on a distance map (for example, nonzero ± several centimeter errors), making it hard to see true construction errors. By applying Gaussian smoothing to the point cloud, random variations are greatly reduced and clear patterns appear on the difference map. For example, if a region of the point cloud is uniformly 3 cm higher than the model, it becomes easier to conclude that the construction was actually 3 cm high in that area. Conversely, small differences below ±1 cm are effectively removed by smoothing and can be considered negligibly small for practical purposes, reducing noise-driven misjudgments in evaluation.

Gaussian kernels are also useful in registration processes between point clouds and models. The commonly used ICP (Iterative Closest Point) algorithm minimizes distances between corresponding points to compute transformations, but noisy point clouds can produce incorrect correspondences and unstable convergence. Smoothed point clouds reduce the influence of outliers via averaging and make it easier to align based on global shapes. In other words, smoothing local irregularities with a Gaussian kernel helps achieve better global matching.

Thus, Gaussian smoothing assists the entire process of comparing and reconciling point clouds and CIM/3D models, enabling more reliable verification. When scanning on site and overlaying design data, smoothing reduces the risk of incorrect judgments caused by noise—such as misinterpreting “this area is slightly overbuilt” or “this slope is gentler than planned.”

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 in AR and point cloud integration. The introduction benefits can be summarized as follows:

• AR overlay stabilization: Noise-induced jitter from point clouds is suppressed, making virtual models appear fixed precisely in real space. This reduces reliance on markers or manual alignment, allowing users to use AR displays without stress.

• High-precision model matching: Smoothed point clouds with global coordinates allow site data and design models to overlap in the same coordinate system with high accuracy. This enables millimeter- to centimeter-level assessment of differences, making AR a reliable tool 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, time-consuming scans can be treated as one smooth model, simplifying post-processing and analysis.

• Increased on-site work efficiency: Point clouds that require no alignment and stable AR displays greatly enhance the efficiency of on-site checks and instructions. For example, displaying the design model in AR during construction reduces surveying and verification workload and helps detect errors early.

Smoothed point cloud data thus significantly enhances AR reliability and accuracy, strongly supporting the bridge between digital and real.

Future of On-site AR Utilizing LRTK × Gaussian Smoothing

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

• Real-time feedback: LRTK can provide centimeter-level positioning in real time. Combining this with fast point cloud smoothing could enable constructing a smooth point cloud model on the fly while holding a device and immediately reflecting it in AR. If surveying and verification are integrated in real time, instantaneous checks and feedback could be performed throughout the construction process.

• Integration with AR glasses: Currently, AR use is mainly on smartphones and tablets, but as AR glasses (smart glasses) become widespread, LRTK positioning data and smoothed point clouds could be linked to glasses so workers can always see high-precision AR hands-free. For example, a BIM model continuously overlaid without drift through the glasses could digitally support tasks that previously relied on skilled visual judgment.

• Cloud sharing and remote assistance: Sharing LRTK-acquired point clouds and AR information via the cloud allows managers and designers in the office to check model alignment on the site’s point cloud model and send instructions. Smoothed point clouds also reduce data size and communication load, making remote AR support realistic.

• New quality-control standards: If high-precision point clouds and AR become common, standards for construction management and inspection could change. For example, tasks that used to rely on string lines or benchmark markings might be done with AR-projected guidelines, and acceptance criteria could be determined by AR-based difference color maps. Stable, smoothed point clouds would be indispensable in such a digital-first approach.

In this way, LRTK’s high-precision positioning and Gaussian smoothing expand the possibilities for AR use on site. AR with assured accuracy and stability will go beyond mere visualization and become an infrastructure for field work.

First Steps Toward Practical Use: Start High-precision Point Cloud Acquisition with LRTK

We have introduced the benefits of integrating advanced point cloud processing and AR technology. To realize these benefits in actual projects, acquiring high-precision point cloud data is essential. With LRTK, absolute-coordinate point cloud measurement that used to require specialized surveying equipment can be started easily with just a smartphone. Preparation and operation are simple, and non-specialist technicians can perform a 3D scan of the site in a short time. Acquired point cloud data can be automatically uploaded to the cloud and used on site for distance and area measurements and model overlays. Gaussian smoothing can be applied as needed to further reduce noise and improve accuracy.

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