Dr. Kam Sanmugalingam
e-mail: kam (at) shan.me
One of the main areas for improvement in location sensor systems (such as Active Bat or GPS) is the less than expected accuracy in real-world deployments. This is due to an incorrect mapping in the relationship between receivers and transmitters, from incorrect (e.g. incomplete) models of the sensors, environment or physical geometry.
The lab's QoSDReAM2/SPIRIT event-driven middleware defines spatial data models, spatial predictates and spatial relations in a quadtree spatial index to efficiently generate sensor-agnostic region overlap events. My research developed a mathematical framework that received low-level Active Bat ranges. It extended QoSDReAM2/SPIRIT, to reduce location errors during extreme conditions in harsh environments, with:
My 1st year project was on improving the realism of simulated location data. These events could then be sent to the middleware, and look just like a real location tag had generated them. To generate the simulated location events, for generic vehicles/people, we merged event-driven simulation, queuing models, velocity/vehicle models, cellular/hybrid automata, et.c. This work (paper) was discussed with colleagues from other universities and with British Airways' Terminal 5 engineers for their research into crowd control and efficient Ground Service Equipment (anything that can't fly, e.g. baggage trucks) deployment.
I was a member of the QoSDREAM group, and my professors were Andy Hopper and George Coulouris. I was also a member of Girton College.