Understanding Scanner Accuracy Specifications

This article explains, from the perspective of surveying error theory, the formal meaning of each scanner accuracy specification, how they differ from one another, and how to interpret them correctly during device selection and project acceptance. It helps distinguish easily confused concepts such as accuracy vs. precision, relative vs. absolute accuracy, and post-processed vs. real-time output accuracy, and explains how test conditions and statistical conventions affect the reported figures.

1. Terminology Overview

The table below summarizes the core accuracy metrics covered in this article, with their formal meaning and the typical wording found in technical specifications. Full definitions are given in the following sections.

2. Accuracy and Precision

2.1 Definitions in Surveying Error Theory

In surveying error theory, observation errors are classified by nature into systematic errors (regular, reducible through calibration or correction) and random errors (also called accidental errors; statistically distributed and not fully eliminable). On this basis, the accuracy-related terms are defined as follows:

• Precision: the degree of dispersion of measurement errors, reflecting random error only, typically quantified by dispersion metrics such as mean square error or standard deviation. In surveying, the closer repeated observations are to one another, the higher the precision.
• Accuracy: the closeness of a measured result to its true value, affected by both systematic and random errors. When systematic error cannot be ignored, the result may deviate from the true value as a whole even if precision is high.
• Trueness: the closeness of the mean of a large number of repeated observations to the true value, mainly reflecting the magnitude of systematic error.

Their relationship can be summarized as: Accuracy = Trueness (small systematic error) + Precision (small random error).

For point cloud data, the more concentrated repeated samples of the same point are, the higher the precision; the closer the measured dimensions and coordinates are to the true values, the higher the accuracy. The two must be evaluated separately and cannot substitute for each other.

2.2 Engineering Significance

• High precision does not mean high accuracy. Small point cloud thickness and highly consistent repeated measurements only indicate small random error; if systematic error exists, the measured dimensions may still deviate from the true value as a whole.
• Some device specifications use precision metrics such as "repeatability" or "point cloud thickness" as the primary accuracy statement, which can easily be misread as accuracy metrics. These should be distinguished carefully during selection and comparison.
• The core of project acceptance is usually accuracy, i.e. how well the measured dimensions and coordinates agree with the true values.

2.3 Nature of the Lixel L3 5 mm Specification

The 5 mm relative accuracy stated for the Lixel L3 is derived by comparing point-to-point distances measured within the point cloud against a baseline (true value) established by a total station. It characterizes "the accuracy of dimensional measurement" and belongs to the accuracy category, not merely a precision metric.

3. Relative Accuracy and Absolute Accuracy

In surveying, accuracy evaluation is divided into internal consistency and external consistency: the former evaluates internal agreement among observations, while the latter evaluates overall accuracy against an independent high-precision external reference (the true value). The relative and absolute accuracy of a point cloud correspond to these two evaluations respectively.

3.1 Relative Accuracy (Internal Consistency)

Definition: the deviation of any point-to-point distance measured within the point cloud from its true value, reflecting the fidelity of the internal geometry; it belongs to internal consistency.

Characteristics:

• Characterizes the fidelity of local dimensions and shapes. For example, if a wall is truly 5.000 m long and the point cloud measures 5.005 m, the 5 mm deviation reflects relative accuracy.
• Does not depend on RTK or ground control points; mainly determined by SLAM mapping quality and the internal geometric self-consistency of the point cloud.

Reference figures: Lixel L3 is 5 mm (approx. 500 m² dedicated test environment); K1 / K2 / L2 Pro are approx. 1 cm.

3.2 Absolute Accuracy (External Consistency)

Definition: the deviation of a point's coordinates in the point cloud from its true coordinates in a given reference frame (such as WGS84, CGCS2000, or a project coordinate system), reflecting the geolocation accuracy of the result; it belongs to external consistency.

Characteristics:

• Characterizes the positioning accuracy of a point in the coordinate reference frame, i.e. the deviation between a point's coordinates (X', Y', Z') in the point cloud and its true coordinates (X, Y, Z).
• Depends on RTK positioning or ground control points to provide an external coordinate reference constraint.

Reference figures: Lixel L3 is 2 cm (RTK fusion); K1 / K2 / L2 Pro are approx. 3 cm.

3.3 Common Misconceptions

Misconception 1: "The L3 accuracy is 5 mm, so the spatial position of every point on the scanned building is accurate to 5 mm."

This confuses relative accuracy with absolute accuracy. 5 mm is the relative accuracy, characterizing the fidelity of internal dimensions; the positioning accuracy of the building in the coordinate reference frame is the absolute accuracy of 2 cm. The two mean different things and are not equivalent.

Misconception 2: "A device rated at 3 mm must outperform the L3."

This conclusion lacks a basis for comparability. A valid comparison requires all three of the following conditions to match:

1. Consistent metric nature (both accuracy or both precision; some 3 mm figures are actually precision metrics such as point cloud thickness);
2. Consistent test conditions (site scale, presence of ground control points, static measurement vs. dynamic capture, etc.);
3. Consistent accuracy type and statistical convention (relative or absolute; 1σ, 2σ, or RMSE).

4. Post-Processed Accuracy and Real-Time Output Accuracy

4.1 Where the Two Types of Accuracy Arise

The processing chain from capture to delivery is as follows:

Capture device → Mobile observation (live preview) → Import to workstation → Software post-processing → Final point cloud

The meaning and applicable scenarios of the accuracy at each stage are shown below:

4.2 Prerequisites for the Nominal Accuracy in Specifications

The accuracy stated in specifications (e.g. the L3's relative 5 mm) refers to post-processed accuracy. For projects requiring 5 mm relative accuracy, the post-processed result must be used as the basis for delivery; real-time output point clouds must not be used as a substitute.

Real-time output accuracy is lower than post-processed accuracy and is mainly used to judge data completeness on site (whether re-capture is needed); it is not the final deliverable.

4.3 Misconception

Misconception: "The point cloud in the on-site live preview has already reached 5 mm accuracy."

This is incorrect. The on-site real-time output has an accuracy of relative 1 cm / absolute 2.5 cm (RMSE); 5 mm is the relative accuracy of the post-processed result. The two must not be confused.

5. Precision Metrics and Attitude Metrics

The metrics in this section fall into two categories of different nature and must be understood separately:

• Precision metrics: repeatability and point cloud thickness, characterizing the dispersion of observations and reflecting the magnitude of random error;
• Attitude metrics: levelness and verticality, characterizing the device's ability to orient to the direction of gravity; their errors are systematic and affect the overall geometric orientation of the result.

The two cannot be converted into each other, nor can they replace accuracy metrics.

5.1 Repeatability

Definition: the degree of agreement among repeated observations of the same quantity under identical conditions (same device, same object, same method, within a short time), usually quantified by the standard deviation or mean square error of the repeated observations.

Notes:

• Repeatability is the quantitative measure of precision under the strictest case of "fully identical observation conditions", reflecting random error only.
• The corresponding concept is reproducibility, i.e. the agreement of results after changing observation conditions (different device, operator, time, or site). Because conditions are relaxed, its dispersion is usually greater than that of repeatability.
• High repeatability only indicates stable results, not results close to the true value. If systematic error exists, repeated observations can consistently deviate from the true value.

Conceptual hierarchy: precision is the superordinate concept, with repeatability and reproducibility being its two specific measures under different observation conditions. The relationship is: under precision (the dispersion of observations) there are repeatability (fully identical conditions) and reproducibility (conditions changed).

Reference figure: Lixel L3 repeatability is 2 cm.

5.2 Point Cloud Thickness

Definition: the dispersion range, along the plane's normal, of points sampled multiple times on the same plane; it belongs to the precision category.

Notes:

• This metric characterizes mapping convergence and noise level, reflecting the magnitude of random error; it does not characterize the accuracy of dimensional measurement.
• The smaller the value, the smaller the dispersion along the plane's normal, and the thinner and cleaner the result appears.
• Small point cloud thickness does not mean the measured dimension of that plane is accurate to the corresponding value.

Relationship with repeatability: both are specific manifestations of precision, differing in the dimension measured.

Reference figures: Lixel L3 point cloud thickness is ≤5 mm; Lixel L2 Pro point cloud thickness is 5 mm.

5.3 Levelness and Verticality

Both derive from the device's ability to measure its attitude relative to the plumb (gravity) direction, but they characterize different objects and must be distinguished.

Levelness

Definition: the deviation of the horizontal reference plane established by the device from the true horizontal plane (the plane perpendicular to the plumb line).

• Characterizes the accuracy of the horizontal reference, affecting the overall horizontal attitude of the point cloud.
• Insufficient levelness tilts the point cloud as a whole, appearing as uneven floors, gradual height differences within the same floor, and layering or misalignment at the joints of multiple data segments.
• Significantly affects elevation measurement, planar drawing, earthwork calculation, etc.

Verticality

Definition: the deviation of vertical structures in the point cloud (such as walls, columns, curtain-wall mullions) from the true plumb direction.

• Characterizes the plumb fidelity of vertical structures, affecting the correctness of facade geometry.
• Insufficient verticality makes walls appear tilted in the point cloud, elevation drawings inconsistent with reality, and the offset at the top of high-rise buildings accumulates and amplifies with height.
• Significantly affects facade surveying, curtain-wall engineering, verticality inspection, etc.

Relationship: levelness and verticality are geometrically complementary and originate from the same inertial attitude measurement system. Once the horizontal reference tilts, the plumb fidelity of vertical structures deviates accordingly, so specifications usually characterize this capability with a single attitude-accuracy metric (labeled as levelness or verticality).

Error amplification with distance: attitude errors appear as angular quantities, and the positional deviation they cause grows linearly with distance, estimated as positional deviation ≈ distance × tan(angular error). Taking 0.005° as an example:

• At 10 m, the corresponding deviation is about 0.9 mm;
• At 100 m, about 8.7 mm;
• At 200 m, about 17.5 mm.

Therefore, in large-scale scenes or high-rise work, attitude accuracy affects the result far more than in small-scale scenes.

Role of RTK fusion: the absolute position constraint provided by RTK suppresses long-term drift of the inertial attitude, significantly improving attitude accuracy. For the L3, it is 0.02° without RTK fusion and improves to 0.005° with RTK fusion.

Reference figures: Lixel L3 levelness is 0.02° (no RTK) / 0.005° (RTK fusion); Lixel L2 Pro levelness is 0.03° (no RTK) / 0.015° (RTK fusion).

5.4 Comparison of the Four Metrics

5.5 Applicable Scenarios

• For projects requiring point cloud cleanliness (e.g. high-precision drawing, 3D result presentation), focus on point cloud thickness.
• For projects requiring result consistency (e.g. periodic re-survey, deformation monitoring), focus on repeatability.
• For projects requiring a horizontal reference and elevation (e.g. floor plans, earthwork calculation), focus on levelness.
• For projects involving vertical structures (e.g. facade surveying, curtain-wall engineering, high-rise verticality inspection), focus on verticality.

6. RTK-Related Accuracy

RTK positioning is the key step by which a result obtains absolute coordinate accuracy.

6.1 Definition of RTK

RTK (Real-Time Kinematic) obtains centimeter-level absolute coordinates in real time through carrier-phase differential solving between a base station and a rover. Its role is to determine the device's absolute position—latitude, longitude, and elevation—in a given reference frame (such as WGS84, CGCS2000).

6.2 Key Figures

The RTK accuracy figures for the Lixel L3 are:

• Horizontal (planar) accuracy: 0.8 cm + 1 ppm
• Vertical (elevation) accuracy: 1.5 cm + 1 ppm
• Maximum antenna gain: 4.0 dBi
• Supported constellations: BDS / GPS / GLONASS / Galileo / QZSS / NavIC / SBAS (more than the standard L2 Pro)

Here "+1 ppm" means the error grows linearly with distance from the base station, increasing by 1 mm per additional km. For example, at 5 km from the base station, the planar accuracy is about 0.8 cm + 5 × 1 mm = 1.3 cm.

6.3 Effect of RTK Fusion State on the Result

For projects requiring absolute coordinates (e.g. geolocation alignment, multi-device result fusion, registration with CAD drawings), RTK should be enabled.

7. Statistical Conventions of Accuracy: 1σ / P68 / P95 / RMSE

The same accuracy figure means different things under different statistical conventions. Clarifying the statistical convention is a prerequisite for comparing accuracy figures and aligning on project acceptance.

7.1 Terminology

• Mean square error: the fundamental accuracy metric in surveying, defined as the positive square root of the arithmetic mean of the squared true errors of the observations; its mathematical form is identical to the root mean square error (RMSE). The smaller the mean square error, the more concentrated the error distribution and the higher the accuracy. In point cloud accuracy evaluation, it is often expressed via the mean square error of same-point residuals or distance residuals.
• Standard deviation (σ): a measure of the dispersion of observation errors, equal to the positive square root of the variance, mainly reflecting random error. When errors approximately follow a normal distribution, the probability that a true error falls within ±1σ is about 68.3%, and within ±2σ about 95.4%. When systematic error is negligible, the mean square error and the standard deviation are approximately equal.
• Percentile (Pxx): a statistical statement that a given proportion of measured-point errors falls within the corresponding range, e.g. P68 means about 68% of measured-point errors are within that range.

7.2 Common Statistical Conventions

7.3 Effect of Statistical Convention on the Figure

The stated figure for the same accuracy level can differ by about a factor of two under different conventions. For the L3, if the relative accuracy is stated under P95, the corresponding figure is about 10 mm rather than the 5 mm under P68.

For comparison and acceptance, follow these principles:

• When a specification does not state σ or a percentile, it is usually understood as the 1σ / P68 convention;
• When an acceptance standard requires the 2σ convention, convert proactively;
• Figures under different percentiles (e.g. P68 vs. P95) cannot be compared directly.

8. Effect of Test Conditions on Accuracy Figures

8.1 Significance of Stating Test Conditions

Accuracy figures depend heavily on test conditions; the L3's note of "500 m² dedicated test environment" states the prerequisite for the figure. Main influencing factors include:

• Site scale: SLAM accumulated error differs between 500 m² and 5000 m² scenes;
• Site geometric features: SLAM stability differs significantly between feature-rich scenes and weak-texture wall scenes;
• Capture path: closed-loop paths and one-way straight paths have different error-accumulation characteristics;
• Presence of ground control points (GCP): GCP-free post-processing vs. GCP-constrained correction can differ by an order of magnitude in accuracy.

The L3's nominal 5 mm relative accuracy holds under the conditions of GCP-free post-processing, a 500 m² composite indoor scene, and the P68 convention.

8.2 Points to Verify When Comparing Accuracy Figures

Because differences in test conditions and statistical conventions significantly affect the figures, when comparing third-party specifications, verify the following four:

1. Test scene: whether it is an ideal scene (e.g. a small, static, enclosed space);
2. Metric nature: whether a precision metric (e.g. point cloud thickness) is used in place of an accuracy metric;
3. Statistical convention: whether a percentile or σ multiple is clearly stated;
4. Measurement method: static measurement or dynamic capture.

When these conditions are unclear, the comparability of the figure is questionable.

9. Summary of Accuracy Figures by XGRIDS Model

The P1 is a spatial camera focused on visual presentation and does not provide engineering measurement accuracy figures.

10. Applying Accuracy Figures in Project Acceptance

10.1 Workflow for Matching Accuracy Requirements

To assess whether a device's accuracy meets project requirements, proceed in three steps:

Step 1: Define the project accuracy requirement

• Accuracy type: relative or absolute;
• Statistical convention: 1σ, 2σ, RMSE, or Max;
• Test conditions: site scale, whether GCPs are deployed, static measurement or dynamic capture.

Step 2: Verify the device figures against the specification

• Align each of the above items one by one; avoid comparing a single figure in isolation.

Step 3: Reserve an accuracy safety margin

• The nominal relative accuracy (e.g. 5 mm) may degrade to 8–10 mm in real projects due to site conditions;
• The nominal absolute accuracy (e.g. 2 cm) may degrade to 3–4 cm at long distances from the RTK base station.

10.2 On-Site Quality Check Points

On-site quality checks can observe the following eight aspects:

1. Layering: check levelness and RTK fusion state;
2. Drift: check whether loop closure is achieved;
3. Misalignment: check the quality of multi-segment data joining;
4. Noise: check point cloud thickness and filtering;
5. Dimensional consistency: spot-check known lengths to verify relative accuracy;
6. Coordinate consistency: compare against known control points to verify absolute accuracy;
7. Color fidelity: check the quality of true-color point cloud coloring;
8. Coverage completeness: check for missed areas or holes.

11. Clarifying Common Misconceptions

Misconception 1: "The L3 accuracy is 5 mm, so every dimension in the result is accurate to 5 mm."

Distinguish relative from absolute accuracy. 5 mm is the relative accuracy, characterizing the fidelity of internal dimensions; the positioning accuracy of a point in the coordinate reference frame is the absolute accuracy of 2 cm.

Misconception 2: "A third-party device rated at 3 mm outperforms the L3."

Compare along the same dimension, i.e. matching statistical convention, test conditions, and metric nature (accuracy or precision). Some figures rated at 3 mm are actually point cloud thickness (precision), which is not in the same dimension as the L3's 5 mm relative accuracy (accuracy).

Misconception 3: "A clean-looking live point cloud on site has already reached 5 mm accuracy."

What is observed on site is the real-time output, whose L3 accuracy is relative 1 cm / absolute 2.5 cm (RMSE); 5 mm is the relative accuracy of the post-processed result.

Misconception 4: "Small point cloud thickness means high accuracy."

Point cloud thickness is a precision metric, reflecting only the dispersion along the plane's normal; it does not characterize the accuracy of dimensional measurement.

Misconception 5: "After enabling RTK, absolute accuracy will necessarily reach 2 cm."

RTK accuracy is also affected by distance from the base station (the 1 ppm term), satellite conditions, occlusion, and multipath. In dense urban high-rise areas, under tree canopies, etc., RTK accuracy degrades noticeably.

Misconception 6: "Levelness of 0.005° is irrelevant to real projects."

Attitude errors are angular quantities, and the positional deviation they cause amplifies linearly with distance. 0.005° corresponds to about 8.7 mm of deviation at 100 m. For facade surveying, curtain walls, high-rise buildings, etc., levelness and verticality directly affect the alignment of structural lines.

Misconception 7: "Repeatability is accuracy; consistent repeated measurements mean an accurate result."

Repeatability belongs to precision, indicating only stable results and small random error. If systematic error exists (e.g. attitude reference bias, scale-factor error), repeated observations will consistently deviate from the true value—high repeatability but low accuracy.

12. Glossary

• Mean square error: the fundamental accuracy metric in surveying, the positive square root of the mean of squared true errors; mathematically identical to RMSE.
• σ (sigma, standard deviation): a measure of the dispersion of observation errors, the positive square root of the variance, mainly reflecting random error.
• P68 / P95 / P99 (percentiles): statistical statements that a given proportion of measured-point errors fall within the corresponding range; P68 approximately corresponds to 1σ, P95 to 2σ.
• RMSE (Root Mean Square Error): the positive square root of the mean of squared differences between measured and true values; called mean square error in surveying.
• Internal consistency / external consistency: respectively, the internal agreement among observations and the agreement between observations and an independent high-precision external reference; corresponding to the relative and absolute accuracy of a point cloud.
• Repeatability: the agreement of repeated observations under fully identical conditions; one of the specific measures of precision.
• Reproducibility: the agreement of repeated observations after conditions change; its dispersion is usually greater than that of repeatability.
• Levelness: the deviation of the horizontal reference plane established by the device from the true horizontal plane.
• Verticality: the deviation of vertical structures in the point cloud from the true plumb direction.
• GCP (Ground Control Point): a known-coordinate point used to constrain the absolute accuracy of a result.
• RTK (Real-Time Kinematic): real-time kinematic carrier-phase differential positioning.
• WGS84 / CGCS2000: coordinate reference systems (the former is the World Geodetic System; the latter is the China Geodetic Coordinate System 2000).
• PPM (parts per million): used to express the error component that grows linearly with distance.
• SLAM (Simultaneous Localization and Mapping): simultaneous localization and mapping.