AI-driven warehouse automation: a unified edge computer vision framework for real-time package tracking and conveyor monitoring

High-speed warehouse automation depends on reliable monitoring of package flow and operational anomalies across sortation conveyors, yet conventional solutions typically deploy separate physical sensors for individual tasks. This paper presents a unified edge computer vision framework that derives multiple conveyor-monitoring functions from a single overhead camera and a shared detection-tracking pipeline. The key novelty is the use of continuous, identity-preserving package trajectory tracking and analysis as a shared basis for virtual sensors : software-defined trigger lines and detection zones. Throughput counting, doubles detection, and reversal detection are derived from the shared trajectories, while dwelling detection reuses the same per-frame detections for a periodic zone-occupancy check; all four functions are therefore served by one detector pass per processed frame, without task-specific sensing hardware or redundant inference pipelines. The framework combines a fine-tuned RF-DETR Small detector with ByteTrack multi-object tracking to generate identity-preserving package trajectories in real time. It was evaluated at an operational sortation facility, where throughput counting achieved 98.05% accuracy; doubles detection achieved 96.0% precision and reversal detection 96.6% precision, while event recall, estimated on approximately six hours of continuously annotated video, was 66.7% for doubles and 88.9% for reversals; and dwelling detection achieved 100% precision and 96.15% recall. These results show that a shared detection and trajectory-analysis backbone can provide high-precision, reconfigurable multi-task conveyor monitoring on edge hardware while supplying trajectory and video metadata unavailable from binary physical sensors, offering a practical building block for AI-driven warehouse automation.