ECE Department Seminar: Indoor Mapping and Localization in a Non-Gaussian World
May 30, 2012
Tin Kam Ho, Piotr Mirowski and Ravi Palaniappan
(in collaboration with Harald Steck, Philip Whiting and Michael MacDonald)
Statistics and Learning Research - Enabling Computing Technologies
Bell Labs, Alcatel-Lucent
Time: 2:00 PM -- 3:00 PM, May 30th (Wednesday), 2012
Location: Babbio Center, Room 203
As mobility moves onto the center stage of telecommunications, there is an increasing need for the network system’s ability to follow users and objects in all types of environments. Radio-Frequency (RF) fingerprinting is an interesting solution for indoor localization and tracking. Accurate localization enables the system to provide location-based services, optimize network deployment and improve the system’s energy efficiency. Our line of research in developing an accurate indoor localization technology has a double focus. On one hand, we have been investigating statistical tools that can leverage the non-Gaussian nature of ambient "noise" for localization. It has resulted in a state-of-the-art technique for WiFi-based positioning, based on probabilistic kernel regression and on comparing the multivariate and non-Gaussian distributions of RF signals with the Kullback-Leibler divergence. We demonstrate localization accuracy of the order of 1m in office environments. On the other hand, we investigated methods to overcome the tedious fingerprinting procedure (i.e., creating signal maps along with precise position information) and repeated calibration that are key to a good performance of any localization system. While trying to automate data acquisition and to create a systematic approach for radio-frequency mapping and fingerprinting, we built a low-cost autonomous and self-localizing robotic platform relying on a Kinect RGB and depth camera. We designed a two-stage localization architecture, performing real-time obstacle-avoidance-based navigation, RGBD visual odometry and localization on an existing map, then Simultaneous Localization and Mapping (SLAM). We compare the applicability of 6-degrees-of-freedom RGB-D SLAM and particle filtering 2D SLAM algorithms for map building and localization in various indoor environment and show that our robot can localize itself while collecting and building WiFi maps in medium-sized office spaces.
Tin Kam Ho is Head of the Statistics and Learning Research Department at Bell Labs. She pioneered research in multiple classifier systems, random decision forests, and data complexity analysis, and pursued many applications. Recently she is working on wireless localization, video surveillance, smart grid data mining, and customer experience management. Her contributions were recognized by the 2008 Pierre Devijver Award for Statistical Pattern Recognition, a Young Scientist Award in Document Analysis and Recognition in 1999, and a Bell Labs President's Gold Award and two Bell Labs Teamwork Awards. She is an elected Fellow of IAPR and IEEE, and served as Editor-in-chief of the journal Pattern Recognition Letters in 2004-2010. She received a Ph.D. in Computer Science from SUNY at Buffalo in 1992.
Piotr Mirowski graduated with a Master's-level engineering degree from Ecole Nationale Superieure in Toulouse, France in 2002, and with an M.Sc. and Ph.D. from the Courant Institute, New York University, in 2007 and 2011, all in Computer Science. His machine learning thesis' subject was "Time Series Modeling with Hidden Variables and Gradient-Based Methods", and his advisor was Prof. Yann LeCun. P.M. joined the Statistics and Learning Research group at Bell Labs in 2011; he also worked at Schlumberger Research in Cambridge, England and in Ridgefield, CT in 2002-2005, and interned at the NYU Medical Center, Google, Standard & Poor's and AT&T Labs during his Ph.D. His areas of specialization are machine learning, image/signal processing and natural language processing; he owns 2 patents, 3 patent publications and authored several articles on applications to geology, epileptic seizure prediction, statistical language modeling and telecommunications.
Ravi Palaniappan is currently a Member of Technical Staff at Alcatel-Lucent Bell Laboratories. His research interests include robotics, sensor networks and high performance computing. He received his M.S and PhD degrees from the University of Central Florida in Electrical Engineering. Previously he worked as a research faculty at the Institute for Simulation & Training in Orlando, FL where he was awarded multi-year project funding from various agencies including NASA, RDECOM and STRICOM. Dr. Palaniappan is an adjunct professor in William Patterson University.