Abstract
<title>Abstract</title> <p>Modern automotive lines build many configurable variants on a shared platform, and every vehicle must match its specification and be free of visible surface defects within a short tact time. Single-camera stations and manual visual checks struggle to meet this combination of variant coverage, throughput, and consistency. We present Automated Vehicle Inspection (AVI), an engineering integration of multi-view machine vision, deep-learning perception, and rule-based decision logic for end-of-line quality control. Eleven synchronized cameras capture a full 360-degree sweep of each vehicle at 3840x2160. Task-specific views are routed to dedicated modules: YOLOv8 for branding and variant-feature detection, EfficientNet for ICE/EV powertrain classification, Gemini 1.5 Flash for mascot OCR, and YOLOv8-Seg for scratch-and-dent segmentation. An evidence-standardization layer converts heterogeneous detections into comparable boolean feature flags; a deterministic rule engine then compares the detected features against a VIN-derived expected manifest to emit an interpretable pass/fail report in about 300 ms. We evaluate on a hybrid corpus: a private OEM multi-view dataset spanning 40 vehicle models, which supports the variant-verification results, and public scratch/dent datasets, which support the surface-defect results. AVI attains 95% end-to-end verification accuracy and 86\% defect-detection recall at 3.3 vehicles/min. Ablations show that both the multi-view design and the dedicated segmentation branch are necessary. The contribution is a deployable pipeline that couples off-the-shelf perception with manifest-conditioned reasoning for variant-aware inspection.</p>