The project focuses on Event Camera based Visual Odometry, mainly for robot localization. Unlike conventional frame cameras that output images based on a fixed frame rate, each pixel of an event camera independently detects relative changes in luminosity and outputs event data asynchronously. It has the advantage of high temporal accuracy, high pixel bandwidth, and responds only to dynamic objects. In order to adapt to challenging environments with different conditions, the work comprises event camera and traditional frame camera binocular visual odometry with the goal to exploit synergies between them. Event-frame visual odometry includes heterogeneous camera calibration techniques. and the project explores the effect on localization due to vision loss in both high and low speed scenarios.
Scholarship holder: Rui Hu.