Pseudo-real-time Wide Area Motion Imagery (WAMI) processing for dynamic feature detection

Real-time information fusion based on WAMI (Wide-Area Motion Imagery), FMV (Full Motion Video), and Text data is highly desired for many mission critical emergency or military applications. However, due to the huge data rate, it is still infeasible to process streaming WAMI in a real-time manner and achieve the goal of online, uninterrupted target tracking. In this paper, a pseudo-real-Time WAMI data stream processing scheme is proposed. Taking advantage of the temporal and spatial locality properties, a divide-and-conquer strategy is adopted to overcome the challenge resulted from the large amount of dynamic data. Each WAMI frame is divided into multiple sub-areas and certain specifically interested sub-areas are assigned to the virtual machines in a container-based cloud computing architecture, which allows dynamic resource provisioning to meet the performance requirement. A prototype has been implemented and the experimental results validate the effectiveness of our approach.