Visual Localization Using Semantic Segmentation and Depth Prediction
This work addresses visual localization for robotics or autonomous systems, though it appears incremental as it builds on existing pipelines with added components.
The paper tackles monocular visual localization by combining semantic segmentation and depth prediction to improve image retrieval ranking and outlier rejection, achieving state-of-the-art performance on the Long-Term Visual Localization benchmark 2020.
In this paper, we propose a monocular visual localization pipeline leveraging semantic and depth cues. We apply semantic consistency evaluation to rank the image retrieval results and a practical clustering technique to reject estimation outliers. In addition, we demonstrate a substantial performance boost achieved with a combination of multiple feature extractors. Furthermore, by using depth prediction with a deep neural network, we show that a significant amount of falsely matched keypoints are identified and eliminated. The proposed pipeline outperforms most of the existing approaches at the Long-Term Visual Localization benchmark 2020.