Research
Hanyu’s research improves the localisation of autonomous platforms using Simultaneous Localisation and Mapping (SLAM), focusing on visual-inertial odometry, which fuses camera images with inertial measurements. Fast motion breaks the small-displacement assumptions behind feature matching and blurs images, sharply reducing the accuracy of state-of-the-art frameworks such as VINS-Fusion. The first part of the work evaluated feature tracking and matching algorithms under high dynamics and optimised the pyramidal optical flow parameters for monocular and stereo configurations. The second develops a lightweight, SLAM-adaptive Transformer model for motion deblurring, supported by a local blur-aware model and IMU priors that give the deblurring stage stronger initial states and cut its processing time.