We are a group at Purdue ECE that innovates at the intersection of computer vision, machine learning, and optics. Specifically, we build visual sensors by jointly inventing cutting-edge computer vision algorithms and optics that synergize to perceive the environment with accuracy, speed, smallness, and power efficiency. Much of our research is inspired by biology. A list of recent and ongoing research threads can be found below.
We are inspired by nature, which contains many examples of vision systems that are incredibly specialized, small, and efficient. Small animals, especially invertebrates, have visual systems comprised of optics and processing algorithms that synergize to excel in specific environments and at specific tasks.
One great example is jumping spiders, which rely on accurate depth perception for predation and navigation. They accomplish depth perception, despite their tiny brains and extremely small power budget, by using specialized optics. In a sequence of papers (ECCV 2016, IJCV 2017, ICCV 2017, PNAS 2019), we designed efficient depth perception algorithms and compact optics that mimic this capability. Parts of the work were recognized by the Best Student Paper Award at ECCV 2016 and the Best Demo Award at ICCP 2018.
Computational visual sensors rely on both optics and algorithms to perform the computation to extract information from the visual input. We establish frameworks that enable the end-to-end optimization of optics and algorithms, which allow the joint consideration of accuracy, computational complexity, sensor complexity, and light efficiency.
In particular, we adopt the metalens as the sample optics to optimize, to leverage its incredible compactness and capability of manipulating light with unprecedented control of angle, spectrum, and polarization (PNAS 2019). This is possible thanks to the recent maturation of metalens fabrication.