XGRIDS Pro Guide™ / Module 5: Project Scale & Map Fusion

5.4 Map Fusion Processing

Creating the fusion project in LixelStudio v4.0, configuring each segment, running the job, and evaluating the result.

Section 1

Hardware Requirements

Map Fusion is more computationally demanding than single-scan processing. Verifying hardware before starting a long fusion job prevents a failure at hour 30 of a 65-hour run.

Recommended specifications for Map Fusion: 16-core processor (AMD Ryzen 9 9950X or equivalent), 64 GB DDR5 RAM with 96 to 128 GB recommended for larger datasets, NVIDIA RTX 3090 with RTX 4090 recommended for optimal performance, and an NVMe SSD with workspace equal to approximately 5 times the raw data size.

Processing time estimates are real. At Medium Quality settings on recommended hardware, expect approximately 20 minutes of compute time per minute of scan data. A 200-minute project is approximately 65 hours of processing. Set Windows power settings to never sleep before starting. Close all non-essential applications. Do not use the machine for other work during processing.

Section 2

Project Setup

Setting up a fusion project correctly at the start prevents configuration errors that require restarting the entire job.

  1. Transfer All Segment Data to the Processing MachineCopy all segment folders from the scanner to a dedicated project folder on an NVMe SSD. Keep each segment in its own folder, named according to your planned naming convention. Verify all files transferred completely before beginning processing.
  2. Open LixelStudio v4.0 and Create a New ProjectSelect Map Fusion as the project type. Do not create a Single Scene project and attempt to add multiple segments. The project type must be set at creation. Single Scene and Map Fusion are separate processing pipelines.
  3. Import Segments in SequenceAdd each segment in collection order. The sequence matters for control point matching; the algorithm processes connections between adjacent segments based on the order they are loaded. Import Segment 1 first, then 2, then 3, in the order they were scanned.
  4. Upload Panoramic Video Files for L2 Pro Panoramic Camera SegmentsIf your segments were scanned with the L2 Pro panoramic camera, upload the matching panoramic video file for each segment at this stage. L2 Pro without panoramic camera and K1 will skip this step automatically. K2 panoramic stitched output workflow compatibility with this step is pending XGRIDS confirmation; for early K2 fusion projects, expect the workflow to mirror the L2 Pro panoramic configuration, but verify the prompt behavior on import.
  5. Verify the Estimated RAM UsageLixelStudio v4.0 displays estimated RAM usage before processing begins. If the estimate approaches or exceeds your installed RAM, enable Low-Memory Mode or reduce the number of segments per fusion job. Running without sufficient RAM causes a failure mid-processing, not at the start.
Section 3

Configuration

The configuration settings for a fusion job control how the algorithm handles SLAM optimization, RTK data, and the alignment computation. The defaults are appropriate for most projects. The settings below are the ones you need to understand when the defaults are not the right choice.

Quality Setting

Fast, Standard, and Slow settings control algorithm depth and iteration count. Standard is the recommended starting point for fusion projects. If the first attempt at Standard fails or produces misalignment, switch to Slow before troubleshooting collection. Fast reduces processing time substantially but at a meaningful accuracy trade; use it only for preliminary verification runs, not final output. For large projects at or above 150 minutes total, use Standard rather than Slow to reduce the risk of memory-related processing failures on machines with 96 GB or less of RAM. Using Standard in this context does not produce bad data; it reduces the algorithm's iteration count while keeping accuracy within professional tolerances.

SLAM Mode

Three options are available. None applies no additional robustness compensation, producing the cleanest optimization and highest accuracy when collection technique was sound and no drift problems exist. It does not disable SLAM processing; it disables the error-compensation algorithms that trade accuracy for resilience. Robust is the default balanced setting. It handles moderate drift and degraded environments better than None at a slight accuracy trade. Narrow Scene is for tunnels and mine galleries specifically; it should not be applied to standard indoor projects.

Loop Closure

Enable Loop Closure when the scan route within any segment included intentional loop paths back to the initialization area. Loop Closure distributes accumulated error across the whole trajectory rather than allowing it to concentrate at the end of the route. If the route was mostly linear with limited loop opportunities, Loop Closure has less effect and can be left at default.

RTK Data Handling

Auto is the default. In Auto mode, LixelStudio applies RTK data where it is valid and ignores it where it is not. Disabled forces the processor to treat all segments as having no RTK. Use Disabled only if RTK data is known to be corrupt or if you intentionally want a relative-coordinates-only output. Do not disable RTK as a troubleshooting step without understanding that doing so removes global coordinates from the entire fused result.

Reference Map Selection

If your fusion project includes segments with different accuracy levels (for example, one segment with excellent RTK and several without), designate the highest-quality segment as the reference map. Other segments align to it. The reference segment should be the one with the most reliable individual SLAM quality and the strongest RTK or control point anchoring.

Section 4

Processing Phases

Map Fusion processing runs in three sequential phases. Understanding what each phase does helps you interpret progress indicators and diagnose where a failure occurred.

  1. Individual SLAM Optimization (Per Segment)Each imported segment runs through the standard SLAM optimization pipeline as if it were a single-scan project. Drift correction, loop closure, and camera-LiDAR fusion happen here. A segment with collection problems may fail this phase; the fusion algorithm cannot proceed until all segments complete their individual optimization.
  2. Inter-Segment AlignmentThe algorithm computes the spatial relationship between every adjacent segment pair using RTK coordinates, matching control point names, or both. This is where the connection method you selected during planning takes effect. Misaligned segments at this phase indicate poor overlap geometry, mismatched control point names, or insufficient shared coverage.
  3. Final Unified Output GenerationThe fully aligned segments are merged into a single trajectory and point cloud dataset. Coordinate system transformation (if RTK is present) and any post-alignment refinements are applied. The output is the single fused point cloud ready for quality review and export.

Do not interrupt processing between phases. LixelStudio v4.0 writes intermediate data during fusion. Interrupting the process mid-run, even between phases, can corrupt the project file and require restarting from import. Set the machine to never sleep and let the job run to completion without interruption.

Section 5

Quality Review

After processing completes, review the fused result before exporting. A quality review that takes 15 minutes catches problems that would waste hours of downstream work in Revit or AutoCAD.

Segment Boundary Inspection

Navigate to each segment boundary, the area where two segments join, and inspect the geometry. A well-aligned fusion shows seamless continuation of geometry across the boundary. Misalignment shows as a step, doubling of surfaces, or geometry that does not line up between the two sides of the boundary. If misalignment is visible, the connection at that boundary needs to be investigated before the result is used.

Accuracy Report and RMSE

Review the accuracy report generated after processing. The RMSE values at each control point connection show how closely the algorithm achieved the theoretical alignment. High RMSE at a specific connection indicates a problem with that segment pair's connection data: poor overlap, a control point with inconsistent position between takes, or an RTK validity issue.

Coverage Completeness

Verify that the fused output includes data from all segments and that there are no unexpected gaps at the transition zones. A segment that failed during individual SLAM optimization may be absent from the fused result; the final dataset will be missing the territory that segment was supposed to cover.

Section 6

Export

The fused result exports from LixelStudio v4.0 as a single dataset. Available formats: E57, LAS/LAZ, RCP, and PLY. The format selection follows the same decision logic as single-scan exports.

E57 filenames must be 20 characters or shorter. LixelStudio's E57 export fails silently on filenames longer than 20 characters; the export appears to complete but the file is corrupt or missing. Rename your project before exporting if needed.

Next: 5.5 Data Management at Scale, managing large datasets after export.

Previous 5.3 Field Collection Next 5.5 Data Management

©2026 Alpine Reality Capture LLC  •  XGRIDS Pro Guide™  •  Site Disclaimer