Answers are grounded in this paper and cite their sources.
We describe a modular video analytics framework for automated staff monitoring in retail and warehouse environments. The system ingests video from multiple cameras and processes each stream through a cascaded pipeline: human detection with joint-level pose estimation, single-camera identity-preserving tracking using Kalman filtering and appearance embeddings, skeleton-driven activity classification via a bidirectional recurrent network, and asynchronous face-based identity resolution. A novel cross-camera re-identification stage projects pixel-space detections onto a shared ground plane through planar homography, clusters co-temporal projections from different views using agglomerative spatial grouping, and links clusters across time with a Hungarian assignment algorithm that jointly optimises spatial proximity and deep appearance similarity. We evaluate the multi-camera pipeline on the EPFL CVLAB multi-person dataset with four synchronised cameras and report qualitative tracking results under varying crowd densities. The dual-lane architecture separates the latency-critical tracking path from the computationally heavier biometric path, and raw twodimensional detections are persisted for retroactive reprocessing when camera calibrations change. We present an honest assessment of the current deployment, noting shortcomings in face recognition accuracy under CCTV conditions and the domain gap introduced by synthetically generated activity training data. Keywords: staff monitoring · multi-camera tracking · person re-identification · homography · activity recognition · pose estimation · cross-view association · surveillance analyticsNote: This manuscript was peer-reviewed and accepted for presentation at the Computer Vision Conference (CVC) 2026, but was withdrawn by the authors due to a lack of institutional funding for publication fees.
No takes yet. Share an insight, caveat, or question.
Alagoni et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: