Real-World Test of SLAM200E + 3DGS | High-Precision Large-Scale 3D Reconstruction

In the field of 3D reconstruction, real-time performance, accuracy, and environmental adaptability remain the core areas of technological breakthrough. The SLAM200E handheld LiDAR scanner, powered by advanced SLAM technology and integrated panoramic imaging, is capable of generating high-precision 3D Gaussian Splatting (3DGS) models.

To validate its performance in real-world conditions, a technical team conducted a large-scale test in a commercial park. The goal was to evaluate the device’s efficiency in data acquisition and its effectiveness in generating 3DGS models under complex environmental conditions.

1. Overview of the SLAM200E 3DGS Case

The test focused on verifying two main aspects: data collection efficiency and quality of model generation. The workflow covered on-site data acquisition, preprocessing, and 3DGS model training. The entire process ran smoothly within a reasonable timeframe, showcasing the scanner’s strong capabilities for efficient 3D reconstruction.

Technical Workflow

General Workflow

The solution follows a structured pipeline: data collection → preprocessing → 3DGS pre-training → model training → output.

Flexible Training Options

For 3DGS pre-training results generated from undistorted imagery, users can select different training tools depending on project requirements. Options include the built-in training module or third-party/self-developed tools, ensuring compatibility across diverse workflows.

In this test, the route marked in red within the workflow diagram was adopted.

Implementation and Data Processing

(1) Site Survey & Route Design

The test area was divided into above-ground and underground sections.

The above-ground section included six high-rise buildings, garden areas, and a small fountain.

The underground section consisted of a ring-shaped corridor connected to restaurants, shops, and a passage leading to a small public park.

Data collection followed a “above-ground first, underground second” principle, with a circular scanning route and repeated coverage in key areas to ensure data completeness.

(2) Data Acquisition

Before scanning, the LiDAR and camera lenses were cleaned to prevent distortion. The SLAM200E and panoramic camera were powered on and synchronized via the SLAM GO mobile app. After confirming synchronization and stability, scanning began along the planned route.

Operators monitored progress in real time through the app, verifying route accuracy and data integrity.

Key Acquisition Notes:

Capture scenes from multiple viewpoints for full coverage.

Use serpentine routes instead of straight paths where possible.

Employ repetitive or saturation scanning in complex/critical areas.

Adjust device height moderately (within ±40° visual difference).

Pay special attention to user-focused details such as pipes, text, sculptures, or other objects of interest.

(3) Data Processing

Point Cloud Mapping

Raw field data was processed using dedicated post-processing software to generate the initial point cloud map.

3DGS Pre-Training

Panoramic images were transformed into undistorted images, with pose data and masks added automatically. This simplified step prepared the dataset for model training.

3DGS Training

At this stage, users may choose from:

Built-in training tools for local computation

Third-party training software

Cloud-based training services for large-scale data processing

This flexibility ensures that users with varying hardware capabilities can achieve reliable 3DGS outputs.

(4) Results and Model Output

The final 3DGS model successfully reconstructed both the above-ground structures and underground spaces of the park. The output supported smooth roaming interactions and delivered high visual fidelity.

Key Advantages Demonstrated:

Efficient Workflow — Streamlined acquisition and processing shorten project cycles

High Precision — Industry-grade SLAM ensures accurate models

Flexible Training Options — Compatible with local, third-party, and cloud workflows

Cost Efficiency — Gaussian Splatting modeling function is permanently open and uses universal data formats, lowering long-term costs

Conclusion

The test results confirm that the SLAM200E handheld LiDAR scanner, with its 3D Gaussian Splatting capability, provides a new benchmark for high-precision large-scale 3D reconstruction. Its combination of real-time data acquisition, model accuracy, and workflow flexibility makes it a powerful tool for applications in urban mapping, facility inspection, underground space modeling, and beyond.

For teams seeking to optimize 3DGS reconstruction efficiency and accuracy, this solution offers a robust and future-ready platform.