GN-Net: The Gauss-Newton Loss for Multi-Weather Relocalization
This addresses the challenge of reliable visual relocalization in varying weather conditions for robotics and autonomous systems, representing a domain-specific incremental improvement.
The paper tackles the problem of direct SLAM methods being sensitive to lighting and weather changes and requiring good initialization by proposing GN-Net, which uses a Gauss-Newton loss to train weather-invariant features for image alignment, resulting in improved robustness and outperforming state-of-the-art methods in experiments.
Direct SLAM methods have shown exceptional performance on odometry tasks. However, they are susceptible to dynamic lighting and weather changes while also suffering from a bad initialization on large baselines. To overcome this, we propose GN-Net: a network optimized with the novel Gauss-Newton loss for training weather invariant deep features, tailored for direct image alignment. Our network can be trained with pixel correspondences between images taken from different sequences. Experiments on both simulated and real-world datasets demonstrate that our approach is more robust against bad initialization, variations in day-time, and weather changes thereby outperforming state-of-the-art direct and indirect methods. Furthermore, we release an evaluation benchmark for relocalization tracking against different types of weather. Our benchmark is available at https://vision.in.tum.de/gn-net.