Recent advances in electronics and sensor design have enabled the development of a camera that can capture hyperspectral datacubes at near video rates. In this talk, I will discuss the problem of tracking objects through challenging conditions, such as rapid illumination and pose changes, occlusions, and in the presence of confusers. Specifically, I will present two state-of-the art trackers (particle filters and mean shift based trackers) to simultaneously track and identify the object based on its reflectance spectra. By exploiting high-resolution spectral features in the visible and near-infrared regimes, it is possible to track objects that appear featureless to the human eye. The combination of spectral detection and motion prediction enables the mean-shift tracker to be robust against abrupt motions, and facilitate fast convergence. Finally, we illustrate the effectiveness of random projections in reducing spectral dimension and improving the computational efficiency. These capabilities are illustrated using experiments conducted on real hyperspectral video data.
This talk is joint with Hien Nguyen, Amit Banerjee, Joshua Broadwater and Philip Burlina.