Rafael Avelar

The emerging field of passive Non-line-of-sight (NLoS) imaging aims to reconstruct a scene hidden from the camera’s field of view by observing indirect light reflected off of a wall. A drawback of some NLoS imaging methodologies in the context of real-life applications is the presence of clutter, or unwanted signals that impede reconstruction attempts of hidden sources. Multispectral analysis combined with signal processing techniques called Blind Source Separation (BSS) have been shown to be a valuable tool in mitigating clutter and improving NLoS reconstructions. These techniques consist of the separation of sources from a set of mixed signals based on their inherent structure or statistical independence. The goal of this project was to expand on previous work with a focus on real large-scale experiments which thoroughly evaluate the performance of different BSS techniques. The methodology involved capturing large-scale NLoS measurements under different variables, such as desired source-to-clutter intensity ratio, number of spectral measurements, and spectral difference between elements. The captured data was analyzed and interpreted with the application of image processing and statistical techniques. Preliminary results indicate that the performance of BSS algorithms improves with a larger desired source-to-clutter intensity ratio and with more spectral measurements. Larger spectral variation in the light from desired sources and clutter also tends to improve the performance of BSS. Further work in the field could involve exploring different types of NLoS imaging algorithms or incorporating thermal NLoS imaging methods that have been shown to have promising capabilities.