Skip to main content

3D Reconstruction of Potted Cotton Plants in a Controlled-Environment Growth Chamber

1. Objective​

  • Establish a high-precision 3D reconstruction pipeline for single potted cotton plants under controlled environmental conditions.
  • Generate standardized image datasets for canopy structural analysis and light-distribution modeling.
  • Evaluate reconstruction accuracy and color consistency under multi-view imaging.

2. Experimental Setup​

2.1 Environment​

  • Location: Controlled-environment growth chamber (adjustable temperature, humidity, and illumination).
  • Background and Floor: Black non-reflective cloth covering both background and floor to suppress unwanted reflections.
  • Lighting:
    • Ambient illuminance: 500–800 lux (uniform diffuse light).
    • Avoid direct or specular lighting.
    • Side-mounted diffused LED panels are recommended; disable ceiling spotlights.

2.2 Rotating Platform​

  • Material: Transparent acrylic turntable (diameter 60–80 cm).
  • Surface Treatment: Covered with a diffuse transparent film to eliminate specular highlights.
  • Markers: Four symmetric reference markers placed on the turntable surface for spatial alignment in Agisoft.
  • Drive: Electrically controlled motorized base, rotating one full revolution in 60–90 seconds at a constant speed.

2.3 Plant Placement​

  • Specimen: Healthy potted cotton plant at vegetative or early reproductive stage.
  • Positioning: Pot center aligned precisely with the turntable center.
  • Stability: Secure with a ring stand or counterweight if necessary.

2.4 Color Calibration​

  • Tool: SpyderCheck24 color calibration chart.
  • Placement: Mounted on the background within the visible field of both cameras.
  • Purpose: Used later for color correction to remove lighting or sensor bias.

3. Data Acquisition​

3.1 Camera Configuration​

  • Devices: Two iPhone Pro cameras (iPhone 13 Pro or later).
  • Angles:
    • Camera A: −45° pitch angle, distance ≈ 1.0 m.
    • Camera B: 0° (horizontal), distance ≈ 1.2 m.
  • Resolution: 4K (3840 × 2160 px) at 30 fps.
  • Focus/Exposure: Manual lock for both.
  • Mode: Continuous video recording.

3.2 Shooting Procedure​

  1. Start the turntable; ensure uniform rotation for one full cycle.
  2. Begin recording simultaneously on both cameras.
  3. Confirm that the SpyderCheck24 and all four markers remain visible throughout rotation.
  4. Stop recording when the turntable completes one revolution.
  5. Rename videos (e.g., Plant01_A_45.mp4, Plant01_B_0.mp4).

4. Pre-Processing​

4.1 Video Frame Extraction and Color Correction​

Using DaVinci Resolve:

  1. Import both videos.
  2. In the Color workspace, calibrate color balance with SpyderCheck24.
  3. Adjust white balance and exposure to maintain natural tones.
  4. Export sequential frames at 1–2-frame intervals (JPG or PNG).

4.2 File Naming Convention​

Use consistent naming for automatic sorting:

Plant01_A_45_####.jpg
Plant01_B_0_####.jpg

5. 3D Reconstruction (Agisoft Metashape Professional)​

5.1 Project Setup​

  • Create a new project and import all images.
  • Group images into two camera sets:
    • Group A (−45° angle)
    • Group B (horizontal view)

5.2 Image Alignment​

Use Align Photos with:

  • Accuracy = High
  • Generic Preselection = Enabled
  • Key point limit = 40000
  • Tie point limit = 10000

Inspect the sparse cloud and verify that all marker points are correctly detected.

5.3 Camera Optimization​

  • Assign marker coordinates (measured or symmetrical).
  • Run Optimize Cameras to refine intrinsic parameters and reduce lens distortion.

5.4 Dense Cloud and Mesh​

  • Build Dense Cloud: Quality = High, Depth Filtering = Mild.
  • Build Mesh: Source = Dense Cloud.
  • Build Texture: Mapping Mode = Generic, Blending Mode = Mosaic.

5.5 Export​

Export the reconstructed model as:

  • OBJ / PLY / GLB (depending on downstream analysis). Include camera positions and coordinate metadata.

6. Post-Processing and Analysis​

  • Color Validation: Compare RGB values of SpyderCheck24 patches to verify calibration.
  • Point-Cloud Cleaning: Use CloudCompare or Open3D to denoise and normalize scale.
  • Phenotypic Trait Extraction:
    • Plant height, canopy width, volume, leaf inclination, etc.
    • Implement with Python + Open3D + NumPy pipelines.

7. Notes​

  1. Avoid any vibration or airflow during recording.
  2. Keep rotation speed constant throughout.
  3. Align camera optical centers with the turntable axis to reduce reconstruction bias.
  4. Maintain consistent EXIF timestamps for all frames.
  5. Save Agisoft project files (.psx) frequently to prevent data loss.
3D_Reconstruction_Cotton/
│
├── Raw_Videos/
│   ├── Plant01_A_45.mp4
│   └── Plant01_B_0.mp4
│
├── Calibrated_Frames/
│   ├── A_45/
│   └── B_0/
│
├── Agisoft_Project/
│   ├── Plant01.psx
│   └── Export/
│       ├── Plant01.obj
│       └── Plant01_texture.jpg
│
└── Metadata/
    ├── Camera_Settings.txt
    └── Turntable_Info.txt

Author: Liangchao Deng, Ph.D. Candidate, Shihezi University / CAS-CEMPS
Experiment conducted in the controlled-environment phenotyping facility.