Calculator D2

2D Blob Analysis for Part Presence & Orientation in Pick-and-Place Applications

Blob analysis is like using digital eyes to find and measure simple shapes—like screws or gears—in a camera image, so a robot knows where to pick them up and how to rotate them.

Industry Applications
Automotive battery module assembly, electronics SMT feeder verification, medical device packaging
Key Standards
ISO/IEC 15426-1 (barcode verifiers), VDI/VDE 2634 Part 2 (machine vision accuracy)
Typical Scale
Field of view: 100 × 100 mm to 400 × 400 mm; pixel resolution: 5–20 µm
Real-Time Requirement
≤33 ms latency from image capture to robot command (30 Hz minimum)

⚠️ Why It Matters

1
Insufficient contrast or noise in camera images
2
Poor blob segmentation and false merges/splits
3
Incorrect centroid or orientation estimation
4
Robot misgrasps or drops parts
5
Increased cycle time and unplanned downtime
6
Reduced OEE and higher scrap rate

📘 Definition

2D blob analysis is a machine vision technique that identifies, segments, and quantifies connected foreground regions (blobs) in binary or grayscale images based on intensity thresholds and morphological criteria. It extracts geometric and photometric properties—including centroid, area, bounding rectangle, orientation angle, and convexity—to support real-time part localization, presence verification, and pose estimation in automated assembly and bin-picking systems.

🎨 Concept Diagram

Camera ImageBlob 1: Washer (θ=0°)Blob 2: Bracket (θ=45°)

AI-generated illustration for visual understanding

💡 Engineering Insight

Blob analysis is not 'just thresholding'—it’s the first link in a deterministic metrology chain. If centroid error exceeds 0.3× gripper jaw clearance or orientation jitter exceeds ½ of the part’s rotational fit tolerance, no downstream calibration or robot path smoothing can recover reliability. Always validate with worst-case physical part stacks—not ideal lab images.

📖 Detailed Explanation

At its core, blob analysis converts a raw image into discrete objects by converting pixel intensities into binary decisions (foreground vs. background) using global or local thresholds. Morphological operations (erosion/dilation) then clean noise and separate touching objects, enabling basic measurements like area and centroid.

Beyond simple segmentation, robust industrial implementations rely on hierarchical validation: first, rejecting blobs outside expected size/orientation bounds; second, applying shape descriptors (circularity, solidity, extent) to discard debris or shadows; third, cross-checking against known part CAD silhouette projections mapped into image space using calibrated homography.

Advanced systems embed blob analysis within a multi-stage vision architecture: coarse blob detection guides ROI selection for sub-pixel edge fitting or template matching; orientation estimates feed into iterative closest point (ICP) alignment for 3D pose refinement; and real-time blob statistics (e.g., count variance across frames) serve as process health indicators—triggering automatic lighting recalibration or camera focus adjustment when drift exceeds statistical control limits.

🔄 Engineering Workflow

Step 1
Step 1: Define part geometry, tolerance stack-up, and gripper kinematic constraints
Step 2
Step 2: Select camera/lens/lighting setup matching field-of-view, depth-of-field, and contrast requirements
Step 3
Step 3: Acquire representative image dataset across part variants, poses, and lighting conditions
Step 4
Step 4: Calibrate camera intrinsics and extrinsics; compute pixel-to-mm mapping at working plane
Step 5
Step 5: Develop and validate blob detection pipeline (threshold → morphology → filtering → feature extraction)
Step 6
Step 6: Integrate pose output (x,y,θ) into robot motion planner via TCP/IP or EtherNet/IP handshake
Step 7
Step 7: Deploy closed-loop verification: post-grasp image confirms orientation before placement

📋 Decision Guide

Rock/Field Condition Recommended Design Action
Highly reflective metal parts under diffuse LED dome lighting Use polarized lighting + polarizing filter; apply adaptive histogram equalization pre-thresholding
Parts with variable surface finish (matte vs. anodized) in same bin Deploy multi-threshold blob analysis with dynamic ROI-based threshold selection per sub-region
Overlapping parts causing merged blobs (>2× nominal part area) Integrate watershed segmentation with distance transform; verify split validity via convexity & aspect ratio constraints
Low Z-height variation but strong X/Y orientation variance (e.g., stamped brackets) Fuse blob orientation with Hough line detection on dominant edge clusters to resolve 180° ambiguity

📊 Key Properties & Parameters

Contrast Ratio

3:1 to 10:1

Ratio of mean intensity of the part region to mean intensity of background region in grayscale image

⚡ Engineering Impact:

Low contrast (<2.5:1) causes unreliable thresholding and false blob detection

Minimum Blob Area

200–2000 pixels (at 5 MP resolution, 10–50 µm/pixel)

Smallest pixel count considered valid for a part-shaped region after binary thresholding and morphological cleanup

⚡ Engineering Impact:

Too low invites noise artifacts; too high misses small or partially occluded parts

Orientation Tolerance

±1.5° to ±8.0°

Maximum allowable angular deviation (in degrees) between measured blob major axis and nominal part orientation

⚡ Engineering Impact:

Exceeding tolerance triggers re-verification or rejects part as misoriented—critical for asymmetric fasteners

Circularity

0.75–0.98 for typical machined parts (e.g., washers, bushings)

Dimensionless metric (4π·Area/Perimeter²) quantifying how closely a blob matches a perfect circle

⚡ Engineering Impact:

Values <0.65 suggest severe occlusion or lighting artifact; >0.99 may indicate specular glare or over-smoothed edge

📐 Key Formulas

Centroid Localization Accuracy

σ_c = √(σ_x² + σ_y²) ≈ 0.35 × √(A_pixel) × p

Estimated RMS centroid uncertainty in mm, where A_pixel is blob area in pixels and p is pixel pitch (mm/pixel)

Variables:
Symbol Name Unit Description
σ_c Centroid Localization Accuracy mm Estimated RMS centroid uncertainty
σ_x Centroid uncertainty in x-direction mm Standard deviation of centroid position in x-axis
σ_y Centroid uncertainty in y-direction mm Standard deviation of centroid position in y-axis
A_pixel Blob area pixels Area of detected blob in pixels
p Pixel pitch mm/pixel Physical size of one pixel
Typical Ranges:
5 MP camera, 15 µm/pixel, 500-pixel blob
0.08–0.12 mm
12 MP camera, 7 µm/pixel, 1200-pixel blob
0.04–0.06 mm
⚠️ σ_c ≤ 0.2 × gripper positional repeatability (typically ≤0.05 mm)

Orientation Uncertainty

σ_θ ≈ (180/π) × (σ_major / L_major)

Approximate standard deviation (degrees) of major-axis angle estimate, where σ_major is length uncertainty of major axis and L_major is its measured length (pixels)

Variables:
Symbol Name Unit Description
σ_θ Orientation Uncertainty degrees Approximate standard deviation of major-axis angle estimate
σ_major Major Axis Length Uncertainty pixels Uncertainty in the measured length of the major axis
L_major Major Axis Length pixels Measured length of the major axis
Typical Ranges:
Well-defined ellipse, L_major = 120 px, σ_major = 1.2 px
0.57°
Noisy edge, L_major = 85 px, σ_major = 2.5 px
1.68°
⚠️ σ_θ ≤ ⅓ of allowable orientation tolerance band

🏭 Engineering Example

Tesla Gigafactory Berlin – Battery Module Line

N/A (applies to manufactured parts)
Circularity
0.87
Contrast Ratio
6.2:1
Minimum Blob Area
480 pixels
Orientation Tolerance
±2.3°
Processing Time per Frame
18 ms @ 60 fps

🏗️ Applications

  • Bin-picking of fasteners in automotive assembly
  • Presence/absence check for PCB components pre-reflow
  • Orientation verification of injection-molded housings

📋 Real Project Case

Vision-Guided Palletizing Robot for Mixed-SKU E-Commerce Fulfillment

Automated distribution center serving Amazon Prime logistics hub

Challenge: Unstructured tote input with random part orientation, variable box dimensions, and reflective surfac...
Vision-Guided Palletizing RobotMixed-SKU E-Commerce FulfillmentLRFOV: 1240×980 mmFOV: 1240×980 mmPolarized DomeUnstructured Tote(random orientation)Dynamic PalletLayer PlanningCNN Pose Estimator(Synthetic Data Trained)Stereo Depth MapReal-time Path RegenGlare (Reflective Surfaces)Contrast ≥ 0.72ChallengeSystemProcessLighting
Read full case study →

🎨 Technical Diagrams

Blob Shape Diversity: Circle (washer), Rotated Rectangle (bracket)
Threshold SweepLowOptimalHighMany false blobsClean separationMissed parts
Major Axisθ = 32°Orientation Angle Extraction

📚 References

[1]
Machine Vision Handbook — Society of Manufacturing Engineers (SME)
[2]
VDI/VDE 2634 Part 2: Optical 3D Measuring Systems — Verein Deutscher Ingenieure
[3]
ISO/IEC TR 29119-4: Software Testing – Part 4: Test Techniques — International Organization for Standardization