LiDAR Point Cloud Processing

LiDAR point cloud processing: from measurement to usable results

Revot Co., Ltd. (株式会社revot) · University of Tsukuba spin-off · Shinjuku, Tokyo

Revot Co., Ltd. is a University of Tsukuba spin-off in Tokyo. Revot provides LiDAR point cloud acquisition and processing, including noise removal, registration, classification, 3D reconstruction, change detection, and volume calculation. Revot handles scanning in GNSS-free environments and accepts projects that involve processing existing point cloud data.

Inquiries in English are accepted by email at contact@revot.jp. The contact form on this site is in Japanese. Revot replies in English or Japanese.

What point cloud data can provide

A point cloud records points in space, each with X, Y, and Z coordinates. LiDAR measures the time between emitting a laser pulse and receiving its reflection. It records surrounding geometry at a density of hundreds of thousands of points per second.

  • Measure dimensions repeatedly from recorded data without returning to the site.
  • Compare measurements over time to identify deformation, deterioration, or accumulation.
  • Calculate the volume of earth or stockpiles.
  • Create 3D models, including digital twins, of equipment and structures.

Point cloud processing turns these coordinates into data for decisions. Revot works backward from the decision the client needs to make to determine the required accuracy and deliverable format.

The point cloud processing workflow

Raw scans require processing before they can support measurement or comparison. Each stage has its own common pitfalls.

StageWorkMain consideration
Noise removalRemove points caused by incorrect reflections, dust, and airborne objects.Excessive removal can erase actual geometry.
RegistrationCombine multiple scans in one coordinate system.Alignment errors can accumulate in spaces with few distinct features.
DownsamplingReduce the number of points.Retain the detail needed for the required accuracy.
ClassificationSeparate ground, walls, equipment, and other objects.Conditions vary by object type.
Meshing and modelingCreate surfaces from points.Hole filling can change the represented shape.
Change detectionExtract differences between measurements.Registration accuracy determines the reliability of detected changes.

Registration affects the quality of the entire result. Inaccurate alignment can make measurement offsets appear to be deterioration or deformation.

Scanning in GNSS-free environments

Indoor spaces, underground areas, tunnels, bridge undersides, and plant interiors often lack usable GNSS signals. Scanning in these environments requires a way to estimate the equipment's position using its own sensors.

SLAM performs localization and mapping together. LiDAR measures surrounding geometry, and changes in those measurements help estimate movement. Revot develops GNSS-free autonomous flight and point cloud acquisition systems for spaces that people have difficulty accessing.

Deliverables and file formats

DeliverableUse
Point cloud dataDimension measurement and import into other systems.
3D models, including meshes and CAD integrationDesign review, interference checks, and digital twins.
Change reportsComparison of deterioration, deformation, and accumulation over time.
Volume and area valuesEarthwork management and inventory assessment.
Orthographic images and cross sectionsReports and communication with stakeholders.

Revot selects the deliverable format according to the intended use. Common formats serve different purposes.

FormatContentsTypical use
LAS / LAZCoordinates, intensity, and classification; LAZ is compressed.Surveying and inspection.
PLYCoordinates and color.Intermediate processing and exchange of colored point clouds.
E57Multiple scans and images together.Raw data from stationary scanners.
PCDPoint cloud data used by robotics libraries.Autonomous navigation and SLAM development.
OBJ / FBX3D models with surfaces.Visualization and presentation; these are not point clouds and are unsuitable for dimensional verification.

Retain LAS / LAZ data with an explicit coordinate system and the original raw data. A processed model alone does not preserve what is needed to repeat the processing.

Revot's development scope and data checks

  • Acquisition systems: Revot develops GNSS-free drones and LiDAR mounting configurations.
  • Processing: Revot registers scans from multiple sessions or devices, removes noise, classifies objects, and reconstructs 3D geometry.
  • Measurement: Revot extracts changes, cross sections, dimensions, volumes, and areas.
  • Models and drawings: Revot creates meshes, reduces data size for browsers and field tablets, and produces CAD floor plans from indoor scans.
  • Camera integration: Revot maps image color onto points and places the positions of cracks on the 3D model.
  • Operational systems: Revot develops systems that incorporate recurring measurements into operations.

Applications include equipment inspection, infrastructure surveys, disaster response, and indoor digitization. Existing data may require additional preparation before it supports these uses.

ProblemCauseRequired action
Software cannot open the data or stops responding.Hundreds of millions of points are loaded together.Divide data into tiles, downsample, or create level-of-detail structures.
Dimensions cannot be trusted.Reference coordinates are missing.Establish dimensional references using ground control points during acquisition. These cannot be added afterward.
Repeated scans do not align.Scans use different coordinate systems.Define shared reference points and register the scans.
Objects cannot be separated from the background.The data lacks classification.Separate ground, structures, vegetation, and noise.
Points have no color.LiDAR captures only geometry.Associate camera images with the points.

Without reference coordinates, dimensions cannot be guaranteed, and measurement must be repeated for inspection or as-built verification. Before commissioning data acquisition, confirm what measurement accuracy, in millimeters, is guaranteed.

Measured results

Revot has published measurements from work on colored point clouds using LiDAR and a camera.

ItemMeasured result
Measurement environmentLivox MID-360, indoor, 309 frames, 170,496 points.
Depth alignmentUsing the same depth model, changing the alignment method from a basic transformation to Gaussian process regression improved RMSE from 1.06 m to 0.455 m.
Accuracy limitLiDAR measurement spread was about 0.40 m, with a 5.4 cm gap to the best method.
LiDAR-camera extrinsic calibrationTargetless calibration improved mean projected edge error from 39.15 px to 20.15 px, about 48.5%. Final confirmation requires a checkerboard measurement.
Points within the camera field of viewWhen shooting while stationary, about 6 to 21% of all points were inside the camera field of view.

These numbers come from one specific environment: Livox MID-360, indoor, 309 frames, and an uncalibrated camera. They do not establish performance in other environments.

Indicative costs and durations

ScopeIndicative costDuration
Point cloud acquisition and processing proof of conceptFrom JPY 980,0001–3 months
3D reconstruction pipelineFrom JPY 4,000,0003–8 months
Measurement through deliverable outputFrom JPY 8,000,0006–12 months

All amounts exclude consumption tax. Costs depend on the point count, number of scans, and required accuracy. Revot provides a cost breakdown in the quotation.

Information to include in an inquiry

Revot accepts inquiries at the idea stage. Include available documents or drawings and the following information.

  • The decision the data must support: deformation, deterioration, accumulation, or dimensions.
  • The measurement location and whether GNSS signals are available.
  • For existing data: format (LAS / LAZ / PLY / E57 / PCD), point count, coordinate system information, and measurement equipment.
  • The required deliverables.
  • The size of the change to detect, in mm.
  • Whether measurement is a single session or a recurring activity.

Revot first checks whether existing data is sufficient or acquisition must be repeated. For comparisons over time, Revot establishes registration references at the start.

Revot accepts inquiries in English by email at contact@revot.jp. The contact form on the site is in Japanese. Revot replies in English or Japanese.

Frequently asked questions

Can Revot process point cloud data that we already have?

Yes. Revot accepts processing-only projects. Revot quotes the cost and delivery time for each project after learning the data volume and required deliverables.

Can Revot scan indoor or underground spaces without GPS?

Yes. Revot uses SLAM for GNSS-free point cloud acquisition and handles autonomous flight in these environments.

Can an indoor drone scan produce cross sections and dimensions?

Yes. Revot can produce 3D models, cross sections, and volume, area, and dimension values. Dimensional guarantees require reference coordinates. Revot confirms the required detectable difference in mm before measurement.

Which is the most important step in point cloud processing?

Registration. Inaccurate alignment makes it difficult to distinguish a change in the object from a measurement offset.

What does noise removal remove?

Revot removes points caused by incorrect reflections, dust, and airborne objects. Revot also classifies and excludes pedestrians, vehicles, and vegetation. Because excessive removal can erase actual geometry, Revot decides which points to keep and which to exclude according to what the client needs to assess.

Do the listed prices include consumption tax?

No. All listed prices exclude consumption tax.

Contact Revot

Inquiries in English are accepted by email at contact@revot.jp; the contact form is in Japanese. Seita Sukisaki, the representative director, replies directly within one to two business days.

Email contact@revot.jp

Revot Co., Ltd. / 3-16 Ichigaya-Tamachi, Shinjuku-ku, Tokyo 162-0843, Japan

Contact form (Japanese)