Multimedia sensor networks have emerged due to the tremendous technological advances in multimedia hardware miniaturization and the application potential they present. However, the time sensitive nature of multimedia data makes them very problematic to handle, especially within constrained environments. In this paper, the authors present a novel approach based on continuous 3D real time reconstruction of the monitored area dedicated for video surveillance applications. Real-time 3D reconstruction allows an important network bandwidth reduction in context to sensor nodes sending descriptive information to the fusion server instead heavy video streams. Each node has to support additional processing in order to extract this descriptive information in real-time, which results in video sensors capturing tasks, data analysis, and extraction of features needed for 3D reconstruction. In this paper, the authors focus on the design and implementation of such sensor node and validate their approach through real experimentations conducted on a real video sensor.
Similarity Graph-Based Camera Tracking for Effective 3D Geometry Reconstruction with Mobile RGB-D Camera