MICRO SCALE · IEEE T-ASE 2025 · SELECTED FOR IEEE IROS 2025

Robotic Concrete Defect Localization on 3D Structural Meshes

Raw inspection video gives civil engineers thousands of disconnected 2D crack snapshots with no global coordinates or metric scale. Developed at the CCNY Robotics Lab, this end-to-end system connects a wall-climbing robot and iPad operator app to AWS serverless CNN segmentation and WebODM 3D point-cloud registration — localizing and measuring cracks directly on full-scale concrete columns and seawalls.

IEEE T-ASE 2025 (Vol. 22) Selected for IEEE IROS 2025 AWS Lambda + WebODM 3D CCNY Robotics Lab
3D reconstructed concrete bridge column with localized and skeletonized structural crack defects
FULL-COLUMN 3D DEFECT RECONSTRUCTION CLICK IMAGE TO ENLARGE
IEEE T-ASE Journal + IROS 2025 Selection Feng, Shang, Hoxha, Hernández, He, Wang, & Xiao (pp. 22324–22336)
2D → 3D Metric Mesh Registration Eliminates duplicate crack counting across overlapping camera frames
Serverless AWS Amplify & Lambda Automated cloud CNN segmentation, mask overlay, and skeleton metrics
iPadOS App Field Operator Interface TestFlight app capturing synchronized RGB, depth, gravity, and GIS pins

ROBOT-TO-CLOUD-TO-3D ARCHITECTURE

From Wall-Climbing Capture to Metric 3D Crack Skeletons

Note on project lineage: this concrete inspection system at the CCNY Robotics Lab (Prof. Jizhong Xiao's group) is a completely independent research line from the Missouri S&T BIRDS steel-girder project (Prof. Genda Chen's group).

01 · CLIMBING ROBOT + iOS RGB + Depth + Gravity GWCR climber & UAV video iPad TestFlight session sync Asset GIS geotagging 02 · AWS SERVERLESS CNN InspectionNet Inference API Gateway + S3 + 4 Lambdas Pixel-level crack masks Auto mask overlay 03 · 3D SfM REGISTRATION WebODM Point Cloud Projects 2D masks onto 3D mesh Global structural frame De-duplicates overlaps 04 · METRIC NDT 3D Crack Skeleton Medial-axis topology True metric length & width Inspector digital twin

FIELD-TO-CLOUD ENGINEERING

iPad Operator App & AWS Serverless Defect Engine

To make robotic defect analytics usable by field inspectors rather than only lab researchers, I built and deployed the cloud-connected mobile workflow:

  • iOS/iPadOS Control & MapView App: Operates the climbing robot via TestFlight, logs RGB + depth + gravity vectors per frame, and pins inspection sessions onto an interactive asset map.
  • 4-Stage AWS Lambda Pipeline: Automatically ingests raw session imagery into Amplify/S3, runs deep CNN crack segmentation, generates visual overlays, and computes dimension metrics.
System architecture linking GWCR climbing robot, iPad MapView app, AWS API Gateway, Amplify S3 storage, and four AWS Lambda inference stages
Deployed field-to-cloud architecture: climbing robot + iPad session storage + AWS API Gateway, S3, and 4 Lambda stages.

3D LOCALIZATION & MEDIAL-AXIS SKELETONIZATION

Turning Pixel Masks into Engineering-Grade Crack Topology

Binary segmentation masks alone cannot tell a structural engineer whether a hairline fracture grew by 4 cm since the last biennial cycle. Our analytics pipeline bridges pixels to structural mechanics:

  • 2D-to-3D Texture & Point-Cloud Fusion: Registers segmented defect masks directly onto a WebODM 3D structure-from-motion model of the entire concrete pier or seawall.
  • Morphological Skeletonization: Extracts 1-pixel medial-axis crack skeletons with homography/depth scale calibration to compute metric crack length and orientation.
  • Subsurface NDT Companion Work: Complemented by automated acoustic impact-echo and rotary-percussive sounding analytics plus thermal IR delamination detection.
IEEE T-ASE 2025 robotic concrete crack segmentation and 3D localization workflow
End-to-end 2D segmentation to 3D structural mesh defect localization (IEEE T-ASE 2025).
Progression from raw concrete image to segmentation mask, binary mask, and 1-pixel topological crack skeleton
Raw concrete surface → CNN segmentation mask → binary mask → medial-axis skeleton for metric length extraction.

PEER-REVIEWED JOURNAL & FLAGSHIP CONFERENCE

Primary Publication

IEEE T-ASE 2025 · Selected for Presentation at IEEE IROS 2025

Robotic Inspection and Data Analytics to Localize and Visualize the Structural Defects of Concrete Infrastructure

Jinglun Feng, Bo Shang, Ejup Hoxha, César Hernández, Yang He, Weihan Wang, and Jizhong Xiao · IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 22324–22336, 2025.

Project Attribution Note

Distinct Institutional Lineage (CCNY vs. Missouri S&T)

This concrete defect localization work was conducted with the CCNY Robotics Lab team (Feng, Shang, Hoxha, Hernández, He, Wang, Xiao) and is completely separate from the Missouri S&T BIRDS steel-girder drone project.