Slide 003: Scan Alignment and Registration¶
Slide Visual¶
Slide Overview¶
This slide covers scan alignment (registration) — the process of transforming multiple individual scans into a unified coordinate system. Students learn target-based alignment, feature-based alignment (ICP), global registration, and how to evaluate alignment quality.
Instruction Notes¶
What Is Registration?¶
Each individual scan is captured in the scanner's local coordinate system — its own X, Y, Z frame of reference. When the object rotates on a turntable or the scanner moves to a new position, each subsequent scan has a different coordinate frame. Registration computes the rigid transformation (rotation + translation) that brings all scans into a common coordinate system, assembling them into a single unified point cloud.
Registration quality directly determines the quality of the final model. Poor registration produces double-walls (two offset copies of the same surface), steps at scan boundaries, and dimensional errors that propagate through all downstream processing.
Alignment Methods¶
1. Target-Based Alignment Adhesive reference targets (coded dots, 3-6 mm diameter) are placed on or near the object before scanning. Each target has a unique coded pattern that the software identifies automatically. When 3+ targets appear in overlapping scans, the software computes the transformation that aligns the target positions. This is the most reliable method for featureless objects.
Target placement guidelines: - Minimum 4 targets visible in each scan (3 required mathematically; 4+ for redundancy) - Distribute targets across the object surface — avoid clustering - Place targets on stable surfaces that will not move between scans - Avoid placing targets on edges or curved surfaces where detection is unreliable
2. Feature-Based Alignment (ICP) Iterative Closest Point (ICP) is an algorithm that aligns two overlapping point clouds by iteratively minimizing the distance between corresponding points. It requires no reference targets but needs: - Sufficient geometric overlap (30-50% shared surface between scans) - Distinctive geometry (not symmetric or featureless) - Reasonable initial alignment (manual rough alignment or turntable angle data)
ICP can fail on symmetric objects (cylinders, spheres), featureless planes, and when overlap is insufficient.
3. Global Registration After pairwise alignment of adjacent scans, global registration simultaneously optimizes all scan positions to minimize accumulated drift error. Without global registration, small alignment errors between sequential scans accumulate — the last scan may not match the first scan even though each adjacent pair was well-aligned. Global optimization distributes the residual error evenly across all scans.
Evaluating Alignment Quality¶
- Alignment residual: Average distance between overlapping points after alignment (target: < 0.05 mm for metrology; < 0.2 mm for general use)
- Visual inspection: Rotate the merged cloud and look for double-walls, steps, or offsets
- Cross-section check: Cut a section through scan overlap regions — well-aligned scans show a single clean surface; misaligned scans show parallel offset surfaces
- Deviation color map: Software can display point-to-point distance as a color map, highlighting problem areas
Troubleshooting Registration Failures¶
| Problem | Likely Cause | Solution |
|---|---|---|
| Scans will not align | Insufficient overlap | Re-scan with smaller angular steps |
| High residual error | Object moved between scans | Re-scan with better fixturing |
| Drift after full rotation | No global registration applied | Run global optimization |
| ICP converges to wrong position | Symmetric object | Add reference targets |
Key Talking Points¶
- Registration is the most critical step between scanning and mesh processing
- Target-based alignment is the most reliable method — always use it for precision work
- ICP is powerful but has failure modes — symmetric and featureless objects need targets
- Global registration eliminates accumulated drift — never skip it for multi-scan datasets
- Always visually verify alignment quality before proceeding to mesh processing
Learning Objectives (Concept Check)¶
- Explain why registration is necessary for multi-scan datasets
- Compare target-based and feature-based (ICP) alignment methods
- Describe the purpose of global registration and when it is needed
- Evaluate alignment quality using residual values and visual inspection
Last Updated: 2026-03-19