Geometry-Aware Depth Anomaly Detection and Adaptive Decision Modeling for Automated Robotic Quality Sorting

Abstract Industrial 3D anomaly detection remains challenging when defective surface structures must be distinguished from normal geometric variations across different industrial components. This study presents a geometry-aware anomaly detection framework that integrates local surface characteristics with global object-shape descriptors for industrial defect identification and localization. The proposed framework is evaluated using the MVTec 3D-AD benchmark, comprising 4,147 industrial 3D scans across ten object categories. Local geometric characteristics, including surface normals, curvature, depth variation, and neighborhood consistency, are combined with global descriptors, including convexity, solidity, and void ratio, to construct a complementary geometric representation. The experimental evaluation examines conventional geometric thresholding, an unsupervised 3D anomaly detection baseline, and the proposed local–global feature-fusion approach. Image-level anomaly detection and pixel-level defect localization are assessed using AUROC, average precision, Intersection over Union, and Dice similarity. Feature-ablation analysis investigates the respective contributions of local and global geometric representations. Furthermore, anomaly severity and prediction uncertainty are incorporated into a three-action quality-inspection strategy comprising acceptance, rejection, and additional inspection. The sorting component is evaluated through a cost-sensitive simulation that considers false acceptance, false rejection, and additional-inspection requirements. The study connects geometric anomaly detection with uncertainty-aware quality control decision modeling and investigates the contribution of complementary geometric information to automated industrial inspection. Physical robotic deployment and validation under actual manufacturing conditions remain outside the scope of this experimental evaluation.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-11703978/v1
Primary Topic
Industrial Vision Systems and Defect Detection
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Geometry-Aware Depth Anomaly Detection and Adaptive Decision Modeling for Automated Robotic Quality Sorting

Shahriar Tasnim
3 links
Research Square
Industrial Vision Systems and Defect Detection
preprint

Geometry-Aware Depth Anomaly Detection and Adaptive Decision Modeling for Automated Robotic Quality Sorting

Shahriar Tasnim
preprint en

Abstract

Abstract Industrial 3D anomaly detection remains challenging when defective surface structures must be distinguished from normal geometric variations across different industrial components. This study presents a geometry-aware anomaly detection framework that integrates local surface characteristics with global object-shape descriptors for industrial defect identification and localization. The proposed framework is evaluated using the MVTec 3D-AD benchmark, comprising 4,147 industrial 3D scans across ten object categories. Local geometric characteristics, including surface normals, curvature, depth variation, and neighborhood consistency, are combined with global descriptors, including convexity, solidity, and void ratio, to construct a complementary geometric representation. The experimental evaluation examines conventional geometric thresholding, an unsupervised 3D anomaly detection baseline, and the proposed local–global feature-fusion approach. Image-level anomaly detection and pixel-level defect localization are assessed using AUROC, average precision, Intersection over Union, and Dice similarity. Feature-ablation analysis investigates the respective contributions of local and global geometric representations. Furthermore, anomaly severity and prediction uncertainty are incorporated into a three-action quality-inspection strategy comprising acceptance, rejection, and additional inspection. The sorting component is evaluated through a cost-sensitive simulation that considers false acceptance, false rejection, and additional-inspection requirements. The study connects geometric anomaly detection with uncertainty-aware quality control decision modeling and investigates the contribution of complementary geometric information to automated industrial inspection. Physical robotic deployment and validation under actual manufacturing conditions remain outside the scope of this experimental evaluation.

Research Square
Lamar University (US)
Industrial Vision Systems and Defect Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon