Learning to Solve Robust Visual Odometry
Robust Visual Odometry (VO) lies at the core of many autonomous driving (AD) systems. Due to the presence of outlying measurements, VO algorithms must be highly robust against outliers. Existing classical approaches usually require solving large-scale non-convex and non-linear least squares problems. Therefore, an algorithm can be trapped at poor local minima. Moreover, optimizing such large-scale problems is also computationally expensive. The objective of this project is to improve the overall performance of existing VO methods by combining classical models with learning-based algorithms to achieve new algorithms with fast convergence while possessing a strong ability to escape poor local minima.
The project is closed: 31/03/2021
This page is only available in english
- Chalmers AI Research Centre (CHAIR) (Centre, Sweden)