iLunch Talk (Nov 13) Role of Prediction Models in Object Tracking  

What: Role of Prediction Models in Object Tracking

When: Friday November 13, 12 noon – 1:00 pm

Where: Zoom Meeting ID: 862 4045 8908  Password: 578745

Who: Suren Kumar
Affiliation: Senior Applied Scientist at Amazon

Abstract: Estimation of a time-varying state such as the position of cars around a self-driving car, measuring a drone’s pose using cameras, the temperature at a location, is heavily reliant on almost perfect sensing. However, almost all sensors have sensing noise or missed observations (GPS sensor in a tunnel). Such imperfection in sensing models can be addressed using priors. In this talk, I will talk about prediction/motion models that complement the sensor information in estimating the state of a time-varying system. I will first show applications of a simple motion-continuity model on a practical and challenging task of estimating object pose. Then, I will talk about the estimation of motion models from sensor data and its application to simultaneous localization and mapping (SLAM) while removing the static world assumption. This talk is based on the research work performed during my PostDoc at the University of Michigan, Ann Arbor.

Bio: Suren Kumar is a Senior Applied Scientist at Amazon and works on multi-label multi-class classification, object detection, and visual search problems. Before Amazon, he was Chief Scientist at a Series-A funded visual search startup. He received his Ph.D. from State University of New York (SUNY) at Buffalo and subsequently pursued a postdoc at the University of Michigan, Ann Arbor.

Host: School of Computing and Information Science, University of Maine

For questions: salimeh.yasaei@maine.edu