Short-Exposure Imaging through Atmospheric Turbulence using Single Photon Image Sensors

点击翻译按钮获取中文摘要

Ground-based long-range passive imaging systems are one of the most critical sensing modalities in civilian and military applications, but they often suffer from distortions due to a turbulent atmosphere. In the presence of atmospheric turbulence, the captured images will become unstable, blurred, and ultimately the details of the objects will be lost. Without any mitigation, the degraded image quality will severely reduce our ability to make informed decisions, identify threat objects, navigate in unknown environments, and perform scientific measurements. Turbulence mitigation methods have been studied for decades, but most optics-based approaches require coherent light sources which are not always feasible, whereas post-processing methods are largely limited to stationary scenes. When an object of interest moves or when the camera is not stationary, both the capturing and restoration of the images remain very challenging tasks. The goal of this project is to develop a new computational imaging framework using the single-photon image sensors, by capturing and recovering moving objects distorted by turbulence. By accomplishing this goal, the project will fill the critical gap in the current imaging technology. This, in turn, will empower new imaging abilities for surveillance, navigation, defense, and remote sensing, all of which are critical to the economy and homeland security.The proposed project will develop methods to acquire very short exposure images, and new algorithms to recover the dynamic scenes. This involves a new type of image sensors that offer state-of-the-art photon resolving capability, which can operate with sufficient signal-to-noise at even extremely short exposures. Four research thrusts will be pursued: (i) to derive the theoretical limits of single-photon imaging through turbulence, by analyzing the photon statistics and its interaction with the atmosphere; (ii) to develop dynamic sampling mechanisms to acquire signals non-uniformly, and to develop optimal exposure controls to enable high-dynamic range imaging; (iii) to develop AI-enhanced image reconstruction algorithms; (iv) to develop an evaluation framework to quantify the performance of image reconstruction algorithms, including building fast turbulence simulators and pattern recognition evaluation schemes. On the education front, the project will create online STEM afterschool programs for 9-12th grade students, focusing on computer vision and machine learning. The emphasis of online content delivery will enable remote teaching when face-to-face meetings are difficult, and allow teaching beyond geographical boundaries.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.