Kirubarajan, Thia (McMaster University)

Purpose

Sensing or inference using data from heterogeneous and geographically-distributed sensors has many civilian as well as defense applications. For example, in defense-oriented surveillance systems, multiple unidentified targets are tracked using noisy data from sensors such as radar, sonar, electro-optical or infrared cameras to identify their locations and courses and classify their types. In civilian urban-monitoring or smart-city systems, multiple cameras are used to monitor traffic and to ensure the safety and security of people in an area. Advances in sensor technologies have resulted in affordable high-quality sensors (e.g., video cameras, acoustic devices, short-range radars) that are ubiquitous around us. Also, unlike before when sensors were usually deployed by those interested in surveillance, now data from ad-hoc sensors-of-opportunity are also available. Admittedly, computing technology has also improved along with advances in sensor technology. However, in order to achieve real-time sensing capability, it is necessary to develop efficient algorithms to process the vast amounts of data from a multitude of sensors (e.g., 4K video data at high framerates, sonar data with extremely high false alarm rates) with a time-varying sensor architecture or configuration. That is, we need algorithms that can adapt to ever-changing target characteristics and sensor configurations at design-time as well as at run-time. We propose to develop multi-level (e.g., at sensor, platform, system and system-of-systems levels) adaptation algorithms to process data from a time-varying set of sensors mounted on platforms that may evolve over time with the objective of accurate tracking, classification and situational awareness. In addition to training highly qualified personnel and advancing the state-of-the-art in sensing technology, the_x000D_