A Robust Real-Time Structure from Motion for Situational Awareness and RSTA

摘要

Maintaining real-time situational awareness of military combat vehicles (manned/unmanned) with onboard vision sensors for either autonomous mobility or reconnaissance missions such as moving target indication (MTI) and automatic target recognition (ATR) while the vehicle is on the move has been technically and operationally challenging. In this paper, we investigate and present a practical implementation of a robust real-time structure from motion technique that allows moving robotic vehicles to be able to reconstruct 3D models from observed 2D features with dynamically adjusting motion parameters. We also demonstrate applications that locate and track moving targets within the structured environment built by the SFM and recognize the targets such as vehicles and humans through a hierarchical shape model.