Key Takeaways
- PTD is the industry standard — Progressive TIN Densification handles 90%+ of points correctly
- Visual correction is surgical — Fix specific errors without affecting correctly classified areas
- Focus on Class 2 (Ground) — All downstream deliverables depend on accurate ground classification
- Check steep slopes and vegetation — These are the most common areas needing refinement
- Integrated tools save time — No export/edit/import cycle needed
Ground classification is the foundation of every terrain model. Master it, and everything built on top—contour lines, volume calculations, cut/fill analysis—achieves survey-grade accuracy.
Modern classification algorithms are sophisticated and handle the vast majority of points correctly. But understanding how they work—and having efficient tools to refine the results—is what separates good deliverables from excellent ones.
This guide explains the PTD algorithm behind automatic classification, shows you where to focus your quality checks, and demonstrates how visual correction tools let you achieve perfect results efficiently.
The Algorithm Behind the Magic
Most modern ground classification tools use some variant of Progressive TIN Densification (PTD). According to research published in Frontiers in Earth Science, PTD is "the most widely employed filtering algorithm because of its robustness and effectiveness in distinguishing ground points from non-ground points."
Here's how it works:
- Initial seeding: The algorithm divides the point cloud into tiles and selects the lowest points in each tile as initial "seed" ground points.
- TIN construction: These seed points form a Triangulated Irregular Network (TIN)—a mesh of triangles representing the initial ground surface estimate.
- Iterative densification: For each triangle in the TIN, the algorithm examines unclassified points that fall within it. A point gets added to the ground class if it meets two criteria: its distance to the TIN surface is below a threshold, AND the angle between the TIN facet and the line connecting the point to the nearest vertex is below a threshold.
- Repeat: The TIN is rebuilt with the newly classified ground points, and the process repeats until no more points qualify.
The elegance of PTD is that it progressively "grows" the ground surface upward from known low points, capturing terrain details while rejecting objects that sit above the surface.
This robust approach works excellently for most terrain types. Let's look at scenarios where a quick visual check can help you achieve perfect results.
Challenging Scenarios to Watch For
PTD handles most terrain excellently, but certain scenarios benefit from extra attention. Knowing these helps you focus your quality checks efficiently. ESRI's documentation describes two categories: "points that should have been classified as ground that weren't (errors of omission), and points that were classified as ground that shouldn't have been (errors of commission)."
Check: Low Vegetation Areas
Vegetation close to the ground surface can pass the distance threshold. In grassy or brushy areas, a quick visual check ensures your DTM represents bare earth rather than vegetation tops.
Check: Steep Slopes
On slopes over 15°, the algorithm may be conservative about including points. If you have steep cut faces or hillsides, verify that ground points weren't excluded—they're easy to add back with the lasso tool.
Check: Bridges and Overpasses
Flat elevated surfaces like bridge decks can be included as ground since roads lead smoothly onto them. If your site has overpasses, a quick check keeps your DTM accurate.
Check: Flat Roofs Near Vegetation
Dense vegetation and flat roofs can sometimes be confused. In mixed urban/vegetated areas, verify buildings and trees are properly distinguished.
Check: Low Outlier Points
Occasional points below ground level (from sensor noise or water penetration) can affect initial seeding. These are easy to spot and reclassify in the viewer.
The Downstream Impact
Why does classification accuracy matter so much?
According to research on DTM creation, "the presence of vegetation and/or slope are the largest contributors to DTM error, causing errors exceeding those caused by instrumental or methodological error. Furthermore, point cloud classification algorithms may also induce DTM errors by misclassifying understory or ground-cover vegetation as ground."
Key insight: In vegetated areas, achieving accurate ground classification has the biggest impact on DTM quality. LiDAR returns from canopy and understory naturally sit above the true ground—so thorough classification ensures your terrain model represents bare earth accurately.
When you achieve accurate ground classification, all downstream deliverables benefit:
- Volume calculations reflect true cut/fill quantities
- Contour lines represent actual terrain elevations
- Drainage analysis correctly identifies low points and flow paths
- Cross-sections show accurate profiles for earthwork estimates
The ASPRS Classification Standard
Before discussing correction tools, it's worth understanding what the classification codes actually mean. The ASPRS standard defines the classification codes used in LAS files:
| Code | Classification |
|---|---|
0 | Never Classified |
1 | Unclassified |
2 | Ground |
3 | Low Vegetation (0.5m - 2.0m above ground) |
4 | Medium Vegetation (2.0m - 5.0m above ground) |
5 | High Vegetation (above 5.0m) |
6 | Building |
7 | Low Point (Noise) |
9 | Water |
17 | Bridge Deck |
Class 2 (Ground) is the critical classification for terrain modeling. Everything else—vegetation heights, building detection, noise removal—depends on having an accurate ground surface first.
The Traditional Correction Workflow
When automatic classification produces errors, the traditional fix involves a multi-software workflow:
- Load the point cloud into CloudCompare or similar software
- Use the segment tool (scissors icon) to manually select points
- For each class you want to create, "manually segment some groups of points representing each class"
- Use 'Edit > Scalar fields > Filter by value' to isolate specific classes
- Apply new classification values
- Export the corrected file
- Re-import into your analysis software
This workflow is powerful but time-consuming. Every round-trip through export/edit/import adds friction and opportunities for error.
Viizor's Approach: PTD + Visual Correction
Viizor Desktop implements ground classification in two stages: automatic PTD processing, followed by visual lasso-based correction.
Stage 1: PTD Classification (Viizor Ground Pro)
The automatic classification uses Progressive TIN Densification with four configurable parameters:
- Max Angle (default: 60°): Maximum slope angle for terrain. Points on steeper slopes may be rejected as non-ground.
- Max Distance (default: 0.05m / 5cm): Maximum height above the TIN surface for a point to qualify as ground.
- Cell Size (default: 50m): Initial tile size for seed point selection. Larger values handle bigger structures but may miss terrain detail.
- Max Iterations (default: 500): Iteration limit for the densification process.
These parameters let you tune the algorithm for your specific terrain. Steep mountainous sites might need a higher Max Angle. Sites with low vegetation might need a tighter Max Distance. Construction sites with large buildings might need a larger Cell Size.
Stage 2: Lasso Reclassification
After automatic classification, visual inspection often reveals errors. Viizor's reclassification tool provides direct correction:
- Activate the reclassify tool from the toolbar
- Draw a lasso polygon around misclassified points using left-click to add vertices
- Right-click to finish the polygon
- Select the correct class from the menu (Ground or Unclassified)
- See results immediately in the viewer
The correction happens visually in 3D. You're not editing a spreadsheet of point IDs or running command-line filters. You see the points, you select them, you fix them.
Practical Correction Scenarios
Scenario 1: Low Bushes in Ground Class
You run automatic classification on a site with scattered shrubs. The DTM looks reasonable at first glance, but closer inspection shows subtle bumps where bushes exist.
Switch to Classification view mode. The ground points (Class 2) show as brown. You notice clusters of brown points sitting above the obvious terrain surface—these are the misclassified bushes.
Draw a lasso around each cluster. Select "Unclassified." The points change class immediately. When you recalculate volumes or regenerate contours, the corrected classification produces accurate results.
Scenario 2: Steep Cut Face Missing Ground Points
A construction site has a steep excavated face. Automatic classification rejected many points on the slope because they exceeded the angle threshold.
In Classification view, you see gaps in the brown ground points along the cut face. The points exist—they're just classified as Unclassified.
Rotate the view to see the slope face clearly. Draw a lasso around the legitimate ground points on the slope. Select "Ground." Your cross-section through the cut now accurately represents the slope geometry.
Scenario 3: Parked Vehicles in Ground Class
A survey captured several parked cars that the algorithm classified as ground (they're low, relatively flat, and connected to the actual ground surface via their tires).
In Classification view, the vehicle rooftops appear as brown ground points slightly elevated above the surrounding pavement.
Lasso each vehicle. Reclassify as Unclassified. Your DTM no longer has phantom bumps where cars were parked during the flight.
Why Visual Correction Matters
The alternative to visual correction is parameter tuning: adjust the algorithm settings and re-run until results improve. But parameter changes have global effects. Tightening Max Distance to reject low vegetation might also reject legitimate ground points in rough terrain. Increasing Max Angle to capture steep slopes might include sloped roofs.
Visual correction is surgical. You fix exactly the points that need fixing without affecting areas where classification worked correctly.
The combination of automatic processing (handling 90%+ of points correctly) and visual correction (fixing the exceptions) produces better results than either approach alone.
The Path to Survey-Grade Accuracy
Every professional knows that thorough review produces the best results. As the experts put it: "Results should be reviewed and refined as needed."
This is simply best practice when working with complex real-world data. The key differentiator is how efficiently you can review and refine the classification.
A workflow that requires exporting to one application, editing in another, and re-importing creates friction that discourages thorough review. A workflow with integrated correction tools makes it practical to inspect and fix issues as part of the normal processing pipeline.
Your DTM is only as accurate as your ground classification. Investing a few minutes in visual review and correction pays dividends in every downstream deliverable.
Try Viizor Desktop Free
Automatic PTD classification + integrated lasso reclassification tools. All processing locally, no cloud upload.
$540 One-time payment
Windows 10/11 • No credit card required for trial