Evaluating Video-to-3D Foundation Models for Wire Harness Assembly Verification

Sammanfattning

Automotive wire harness assembly is a heavily manual process, and a critical part ofthis workflow is Quality Control (QC). Automating this visual verification is challenging:traditional 2D computer vision struggles with spatial occlusions, and lacksessential geometric depth, whereas 3D data can capture a more comprehensive representationof the object. However, existing 3D machine learning methods dependon rigid scanning hardware and extensive training data. To bypass these technicalbottlenecks and reduce the industry’s critical reliance on human labor, this thesisevaluates emerging feed-forward "Video-to-3D" foundation models for automated assemblyverification.Using a custom dataset of smartphone videos, four state-of-the-art (SOTA) architectures(Pi3-X, VGGT, Depth Anything V3, and Map Anything) were evaluated forgeometric fidelity and automated component classification. Results identify DepthAnything V3 demonstrated superior global spatial coherence and strong baselinecounting capabilities, though high false-positive rates across the models indicatefurther refinement is needed. Pi3-x also proved to be a reliable architecture withcompetitive classification performance. Although current models lack the geometricprecision for highly granular verification tasks, often exhibiting wire duplicationor loss of detailed structural definition, they show high promise on broader classificationtasks, such as distinguishing and counting clips versus terminal housings.While fully autonomous inspection requires further refinement to mitigate high falsepositiverates, this flexible 3D approach establishes a step toward modernizing visualquality control of wire harnesses