CT Bag Generator

The grid determines slice dimensions (Width x Height) and slice count (Depth). Larger grids produce finer detail but take longer to voxelize and create bigger files.

Spacing is calculated from: spacing = scene_extent / grid_dimension. The scene extent is determined by the tray size plus a small padding margin.

TSA-Approved Checkpoint CT Scanner Reference

These are real scanners deployed at TSA checkpoints. Image parameters are approximate (vendors don't publish exact specs publicly).

Common characteristics across all TSA checkpoint CT scanners:

• Tunnel opening: ~620mm W x 420mm H (standardized for tray compatibility)
• Dual-energy capable: 80kV / 140kV (for material discrimination)
• Spiral/helical acquisition
• 16-bit pixel data (uint16), MONOCHROME2
• DICOS CT Image output (SOP Class 1.2.840.10008.5.1.4.1.1.2)
• DICOS TDR output for automated threat detection
• Belt speed: ~0.2 m/s (tray passes through in ~3 seconds)

When the user asks for a specific scanner model, use its parameters. When they ask for "realistic" or "production quality", use Scanner-match (512x512x375).

Prerequisites

Requires Blender MCP - a Model Context Protocol server that connects Claude to Blender. See references/blender-mcp-setup.md for full install instructions (download addon.py, install uv, configure ~/.claude.json, activate in Blender).

All scene creation uses mcp__blender__execute_blender_code and mcp__blender__get_viewport_screenshot.

Workflow

1. Choose a Scenario

Pick or randomize a traveler profile. Each produces different bag types and contents.

A second tray for loose personal items is common: watch, belt, phone, wallet, shoes.

2. Create Screening Tray

Default: standard flat tray (660 x 420 x 75mm). See references/tray-specs.md for all sizes and geometry.

The tray is solid. Nothing may overlap or intersect with it. This is the most important constraint in the entire workflow.

• Items sit on tray floor at Z=0.005m (4mm floor + 1mm gap)
• All items must stay within tray interior XY bounds (±320mm X, ±200mm Y) with 5mm clearance
• Items may extend ABOVE the tray rim — that is normal
• No stacking bags on bags. If multiple containers don't fit side-by-side, downsize them until they do
• Gaps between items in the tray are fine
• During voxelization, tray voxels are never overwritten

Gravity rule: Every object must sit on something — the tray floor or another object's top surface. Nothing floats. Items inside bags rest on the bag floor or on items below them. Track running Z height as you pack bottom-up.

After every placement step, validate bounding boxes against the tray interior.

3. Create Container

See references/container-types.md for geometry and hardware per container type.

Containers are hollow shells in the CT volume (2-voxel thick). The shell density (3000) makes the bag outline visible; interior is filled with contents or air.

4. Pack Contents with Randomness

See references/density-catalog.md for the full item catalog.

Randomization rules - simulate how real people pack:

• Vary item count: use random.randint(min, max) for each category
• Offset positions: add random.uniform(-0.02, 0.02) jitter to X/Y placement
• Rotate items: add random.uniform(-15, 15) degree tilt to folded clothes
• Skip categories: not every bag has shoes, food, or a hoodie
• Mix folding styles: some items rolled, some folded flat, some crumpled (scaled unevenly)
• Overlap slightly: real packed clothes compress against each other
• Stuff gaps: socks and underwear get tucked into corners and shoe cavities

Packing is bottom-up. Track a running Z height as items stack. Heavier/flatter items at the bottom, bulky/light items on top. Electronics go in designated pockets (back of bag for laptops, front pocket for tablets/phones).

5. Add Loose Tray Items (Optional)

Items that go directly in the tray, NOT inside a bag:

Place loose items with random rotation on the tray floor. They must not overlap each other or the bag.

6. Constrain All Items

After placing everything:

1. Clamp bag contents inside bag bounds (8mm inset margin)
2. Clamp loose tray items inside tray bounds (8mm inset margin)
3. Verify NO object penetrates the tray shell from below
4. Verify bag shell does not overlap tray shell

7. Apply CT Materials

See density-catalog.md for the full material table. Principled BSDF + emission per tier. Bag shell and tray are opaque (alpha=1.0).

8. Add Threat Items (Optional)

See references/threat-items.md for the full TSA prohibited items catalog with Blender geometry and CT signatures.

Threat items use DICOS Threat Categories per NEMA IIC 1 v04-2023: METAL (firearms, knives), EXPLOSIVE (IEDs, detonators), CONTRABAND (tools, weapons, sporting goods), ANOMALY (suspicious configurations).

Threat items are placed inside bags like any other content during Blender scene creation. No special handling is needed during the Blender sketching phase — threat bounding boxes are computed automatically at DICOS export time (step 11).

9. Voxelize (Optional)

See references/voxelization.md.

Critical voxelization rule: tray mask is built first, then bag and all other objects exclude tray voxels. This prevents interleaving.

Also export a threats.json sidecar with bounding box coordinates for any threat items placed in step 8.

10. Export to DICOS (Optional)

See references/dicos-export.md for the complete conversion workflow with Go code examples.

Uses github.com/jpfielding/dicos.go to write:

• CT Image (.dcs): Multi-frame volume, one axial slice per frame, with ImagePlane geometry, scanner parameters, and HU rescale
• TDR (.dcs): Threat Detection Report with PTO bounding boxes referencing the CT Image, if threats are present

A bundled converter is included at scripts/voxel2dicos/. Copy it to the project and run:

go run ./scripts/voxel2dicos/ tmp/voxels.raw tmp/dicos/

The second argument is an output directory. The script creates:

tmp/dicos/
├── ct.dcs    — multi-frame CT Image
└── tdr.dcs   — Threat Detection Report (only if threats.json exists)

It reads the raw binary + threats.json sidecar, builds a DICOS CTImage and TDR with proper modules, and writes both into the output directory.

11. Generate TDR with Threat Boxes

Always generate a TDR when the scene contains any TSA prohibited items. This is not optional — if banned items are present, a TDR must be written.

See references/tdr-workflow.md for DICOS TDR specifics.

At DICOS export time, scan the Blender scene for any CT_Knife_*, CT_Gun_*, CT_Ammo_*, CT_BoxCutter*, CT_Scissors*, CT_MultiTool*, CT_Firearm*, CT_PipeBomb*, CT_Detonator*, CT_Dynamite*, CT_PlasticExplosive*, or other threat-prefixed objects. For each threat item:

1. Compute its composite world-space bounding box (union of all parts)
2. Convert to volume-relative mm coordinates
3. Add a PTO entry with Threat Category, Assessment Flag (HIGH_THREAT), and bounding box

The TDR is a separate .dcs file that references the CT image. It contains one PTO per threat item, with Alarm Decision set to ALARM.

Naming Convention

• Tray: ScreeningTray
• Container: CarryOnBag, Backpack, Purse, DuffelBag, LaptopBag, MessengerBag
• Bag contents: CT_ prefix (e.g., CT_Laptop, CT_Jeans_Folded)
• Loose tray items: Tray_ prefix (e.g., Tray_Watch, Tray_Phone, Tray_Belt)
• Hardware sub-parts: parent name + suffix (e.g., CT_LaptopBattery, CT_HoodieZipper)