Edge-Assisted Multi-camera Tracking for Digital Twin Systems

The digital twin system combines physical entities or processes with their digital representations, which are marked by real-time data acquisition, processing, and simulation. Object tracking, as a fundamental service to map the physical and digital entities, faces new challenges in the digital twin systems: positioning in 3D space can be affected by object blocking, multi-camera conflicts, and lack of computing resources for real time restore of the object positions in the digital space. To address the challenge, we propose an edge-assisted multi-camera tracking (eMT) approach, which consists of three building blocks: 1) We devise a position calibration scheme to aggregates the 2D coordinates from multiple cameras into unified and accurate 3D coordinates; 2) To deal with the impact of camera view blocking, we propose an adaptive method to adjust the detection boxes according spatial-temporal box traces; 3) To meet the real-time requirements, we utilize edge computing and adopt a person re-identification algorithm based on the 3D coordinates and appearance features. We implemented eMT and embed it into a real digital twin system for real time indoor monitoring. Both trace-driven and real experiment results show that eMT can effectively and accurately restore the 3D coordinates at a low latency for the digital twin systems.