Networked Sensing Systems for Urban Target Recognition
Recent advances in sensing, computing and communication technologies hold the potential for development of a new generation of network-centric target recognition systems. Future advances in sensor integration technology will make these sensor systems cheaper, smaller and capable of deployment in large numbers for improved persistent surveillance in complex military operations. Unattended ground sensor systems (UGS) such as REMBASS, TRSS and OASIS, can collect signals in multiple modalities, including optical/infrared (EO/IR) video imagery, acoustic, seismic and electromagnetic signals. Organic air vehicles (OAVs) and unamnned ground sensor vehicles equipped with EO/IR and electromagnetic snesors are integral parts of the military’s battlefields systems architecture.
While such network systems have demonstrated their utility for target recognition in non-urban environments, extensions to missions in urban scenarios involving large numbers of sensors pose formidable technical challenges. Sensing systems in urban areas must deal with cluttered and obstructed environments, signal blockage by buildings, interfering signals, and the overall density of vehicles, personnel, and manmade structures. Sensing and communications operations are unreliable, and susceptible to uncertain obscuration and fading in such environments. In order to achieve effective, robust target recognition performance, it is essential to deploy redundant, spatially distributed, ad-hoc networked, limited power sensing systems with complementary multi-modal sensing capabilities to create independent information channels that will not be degraded simultaneously.
This raises the fundamental question for our effort: design efficient near-optimal algorithms using this ad hoc network of multi-modal sensors to generate robust, persistent target recognition in urban environments. Specifically, this entails tackling two interrelated major basic research issues:
1) Tributed Information Fusion: How does a distributed large-scale multi-modality sensing system involving diverse imaging and non-imaging sensors fuse information from multiple modalities and sensors over space and time to generate robust, accurate target recognition in the presesnce of highly dynamic, uncertain clutter, sensing and communication conditions?
2) Information Dissemination over Ad Hoc Networks: How will information be transferred among sensors in a power and bandwidth efficient manner that provides the needed quality of service for robust target recognition with minimal resource use?
In order to address the above questions, this proposal outlines an ambitious corss-disciplinary research agenda to develop innovative theoretical frameworks and corresponding algorithms for robust, persistent automatic target recognition using ad hoc networks of heterogeneous, multi-modal imaging and non-imaging sensors in highly dynamic, uncertain urban environments. The motivating themes permeating this integrated approach are distributed, scalable, robust and energy efficient target recognition with large numbers of heterogeneous, multi-modal sensors in uncertain, dynamic environments.
Principal Investigator : Venkatesh Saligrama