Talk by Shintaro Shiba
 |
Shintaro Shiba, Keio Institute |
Bio:
Shintaro Shiba is a Ph.D. candidate at the Department of Electronics and Electrical Engineering, Keio University, Japan. He has been visiting the RIP group (Prof. Guillermo Gallego) at TU Berlin for two years and co-supervised by Prof. Yoshimitsu Aoki (Keio) and Prof. Gallego. Previously, he studied Bachelor’s and Master’s in behavioral neuroscience of songbirds at the University of Tokyo, worked as a machine-learning software engineer for autonomous vehicles at Toyota Motor Corporation. His research interests are event-based vision, computer vision, machine learning, and robotics.
Abstract:
Estimating motion from image sensors is a fundamental problem in computer vision and robotics. Event cameras are novel bio-inspired sensors that provide a signal suitable for estimating motion because their pixels naturally respond to intensity changes, which are tightly related to motions in the scene. However, event data is fundamentally different from conventional frame data, which leads us to rethink visual processing. In this talk, we focus on a single problem setting: motion estimation using an event camera, where we show an example where such a rethinking plays an important role. Furthermore, we demonstrate how the estimated motion can be further utilized for various downstream tasks, such as depth estimation, motion segmentation, intensity reconstruction, and imaging fluctuations of air density. We hope to deepen the understanding of various motion estimation tasks in the emerging field of event-based vision.