Image Processing for Time-Resolved and Very Low-Light Image Sensors
My research is mainly about image processing for Time-Resolved Single Photon Avalanche Diodes (TR-SPAD) cameras. SPAD are photodetectors able to detect the impinge of single photons. In addition, TR-SPAD sensors timestamps the photon arrivals with a picoseconds precision. Moreover, the pixel of a TR-SPAD camera can be asynchronously readout, yielding a stream of asynchronous events. For these characteristics, TR-SPAD data are fundamentally different from the data generated by traditional cameras. In my research, I am investigating novel ad-hoc algorithms for TR-SPAD data. My research plan consists of the following main directions: ;
- ;
- Noise modeling. I will study through an experimental approach the noise affecting TR-SPAD acquisitions. I will statistically characterize all the relevant sources of noise, including shot noise, dark counts, crosstalk and readout noise. ;
- Reconstruction/Enhancement. I will study novel methods for reconstructing digital images from raw TR-SPAD streams of photon detections. Particularly in low-light scenarios, where the stream is strongly affected by noise, the algorithms should implicitly perform a denosing/enhancement step. I plan to investigate both prior-based and learning-based solutions. ;
- Visual Recognition. I will study ad-hoc methods for the purpose of recognition of photon streams, avoiding the unnecessary reconstruction step. I plan to investigate deep learning methods suitable for streams of events. ;
- Change Detections for TR-SPAD streams. I will study change detection methods in TR-SPAD acquisitions to analyze streams of inter photon arrival times, in order to identify discontinuities or changes in the scene. The ultimate goal is to design adaptive algorithms for image reconstruction or processing, exploiting the asynchronous nature of photon arrivals ;
Back to Current Students