MOTION++: A Unified System for Complete Exterior 3D Inspection of a Vehicle

This paper introduces MOTION++, a unified vehicle inspection framework that integrates 3D Gaussian Splatting with RGB-D data fusion to enable high-fidelity reconstructions and robust defect detection, overcoming the limitations of traditional Structure from Motion (SfM) techniques. A core system of this approach is MotBox, which facilitates real-time data preprocessing and manages edge–cloud collaboration, optimizing computational efficiency while ensuring low-latency processing. MOTION++ can generate previously unseen perspectives of a vehicle, significantly enhancing inspection coverage and accuracy. Furthermore, an advanced filtering and consistency-checking mechanism minimizes false positives, reducing the need for manual verification. To ensure efficient resource management and scalability, MOTION++ integrates with the aerOS Meta-OS, which dynamically balances computational workloads between edge and cloud environments. The High-Level Orchestrator (HLO) determines whether 3D Gaussian Splatting model training takes place locally or in the cloud, optimizing resource utilization. Once trained, a lightweight model version is deployed to MotBox for real-time processing and user interaction. When additional computational power is needed, MotBox seamlessly offloads tasks to the cloud for high-resolution synthesis, with aerOS handling the data transfer and task distribution.