Containerized Camera Sensors with Performance Evaluation Toward Distributed 3D Sensing

Photographic 3D sensing is a key technology for building smart spaces. To achieve sufficient localization accuracy and spatial coverage, collaborative multi-camera setups are required. Additionally, high-frame-rate imaging is necessary for tracking fast-moving objects. In this work, we propose a distributed photographic 3D localization framework that utilizes a container-based architecture and collaborates through the MQTT communication protocol. We developed a modular camera sensing unit for real-time 3D shuttlecock tracking in badminton, demonstrating its potential as a prototype edge device for object detection. Each containerized sensing unit consists of three modules: (1) an MQTT agent module for control and data exchange, (2) a camera module for video capture and processing, and (3) a sensing module for object detection. The 3D localization computation is decoupled from individual units and delegated to a content analysis container via MQTT messaging, while an application container is responsible for visualization and application logic. With system-level optimization, our prototype supports eight 120 FPS video streams on a single host (Intel i7-13700 CPU / NVIDIA RTX 4090 GPU), utilizing only $60 \%$ of the available computational resources to achieve real-time 3D shuttlecock localization. To further demonstrate the flexibility and edge deployability of the proposed system, we implemented a distributed version composed of three mobile computing nodes.