Toward computer vision on a tight budget: Todd Zickler, Harvard
- 1:00 pm on Monday, April 29, 2013
- 2:00 pm on Monday, April 29, 2013
- MCS 148
Abstract: At least in the near term, micro-scale platforms like micro air vehicles and micro sensor nodes are unlikely to have power, volume, or mass budgets to support conventional imaging and post-capture processing for visual tasks like detection and tracking. These budgets are severe enough that even common computations, such as large matrix manipulations and convolutions, are difficult or impossible. To help overcome this, we are considering sensor designs that allow some components of scene analysis to happen optically, before light strikes the sensor. I will present and analyze one class of designs in this talk. These sensors reduce power requirements through template-based optical convolution, and they enable a wide field-of-view within a small form. I will describe the trade-offs between field-of-view, volume, and mass in these sensors, and I will describe our initial efforts toward choosing effective templates. I will also show examples of milli-scale prototypes for simple computer vision tasks such as locating edges, tracking targets, and detecting faces. Related publications: - Koppal, et al., "Towards wide-angle micro vision sensors." PAMI 2013. http://vision.seas.harvard.edu/papers/WideAngle_PAMI2013.pdf - Gkioulekas and Zickler, "Dimensionality reduction using the sparse linear model." NIPS 2011. http://people.seas.harvard.edu/~igkiou/sparseldr/nips11.pdf Bio: Todd Zickler received the B.Eng. degree in honours electrical engineering from McGill University in 1996 and the Ph.D. degree in electrical engineering from Yale University in 2004 under the direction of Peter Belhumeur. He joined the School of Engineering and Applied Sciences, Harvard University, as an assistant professor in 2004 and was appointed Professor of Electrical Engineering and Computer Science in 2011. He spent the 2011-2012 academic year as visiting scientist at the Weizmann Institute of Science with support from a Feinberg Foundation Fellowship. Todd's research is focused on modeling the interaction between light and materials, and developing systems to extract scene information from visual data. His work is motivated by applications in photography; face, object, and scene recognition; image-based rendering; image retrieval; image and video compression; robotics; and human-computer interfaces. Todd is a recipient of the US National Science Foundation Career Award and a Research Fellowship from the Alfred P. Sloan Foundation.