Abstract:
A new deep visual odometry (VO) method that selects memory and refines poses considers global information. Existing learning-based methods treat VO as a pure tracking problem by recovering camera poses from image snippets, resulting in severe error accumulation. Global knowledge reduces errors. End-to-end system data preservation is difficult. We design an adaptive memory module that saves information from local to global in a neural memory analog to process long-term dependency.
A module refines previous results using global memory information. Based on feature domain co-visibility, we select features for each view using spatial-temporal attention and previous outputs. Our Tracking, Remembering, and Refining architecture goes beyond tracking. Our approach outperforms state-of-the-art methods by large margins and competes with classic approaches in regular scenes in KITTI and TUM-RGBD datasets. In texture-less regions and abrupt motions, where classic algorithms fail, our model performs well.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
