When Images Aren't RGB - a Light Walk Through AI Visible-Thermal Research
Conventional computer vision applications usually learn features from the visible spectra. From typical image classification to even generative algorithms, RGB imagery is frequently used as training data. In this talk, I'll walk you through some of the research I've worked on from classic image registration to generative modeling in the thermal spectra. Thermal imagery, particularly captured in Long-Wave Infrared (LWIR), visualizes radiated heat from a longer waveform compared to the visible spectrum. As a result, conventional methods from face detection to image alignment cannot be easily transferred from the visible to thermal domain due to differences in intensity and texture. Join me as I touch on these projects and introduce you to some of the motivations, as to why thermal-based computer vision is relevant in our society.
Catherine Ordun is an Executive Advisor of AI at Booz Allen Hamilton, currently working on her dissertation as a PhD Candidate at the University of Maryland Baltimore County. Her research focuses on visible-thermal image translation and registration, in addition to multimodal pain detection. At Booz Allen, she wears many hats from leading AI research for the NIH to technical validation of partner companies through our VC fund, to development of novel algorithms for multiple Federal agencies. She currently has a MPH and MBA, and is looking forward to finally defending her dissertation in October 2023. To escape the dredge of PhD life, she's completing her first scifi book.