Manhattan and Atlanta worlds hold for the structured scenes with vertical and horizontal dominant directions (DDs). To describe the scenes with additional sloping DDs, a mixture of Manhattan worlds seems plausible, but leads to unaligned and unrelated DDs. In contrast, we propose a novel structural model called Hong Kong world. It is more general than Manhattan and Atlanta worlds by considering the sloping DDs. It is more compact and accurate than a mixture of Manhattan worlds by enforcing the orthogonality between vertical, horizontal, and sloping DDs. We further leverage the regularity of Hong Kong world for the line-based SLAM. First, based on a novel consensus voting strategy, we estimate DDs in a semi-searching way. This method is the first one that can determine the number of DDs and achieve quasi-global optimality. Second, we compute the camera pose by exploiting the spatial relations between DDs. This method is accurate and efficient thanks to concise polynomials. Third, we refine DDs by a filter-based method, and use these DDs to improve SLAM accuracy and robustness. In addition, we establish the first dataset of sequential images in Hong Kong world. Experiments showed that our approach outperforms state-of-the-art methods in terms of accuracy and/or efficiency.
(Updated on 20 Dec 2023) We have uploaded dataset to the OneDrive. [here] is the download link. We will provide a "readme" file of this dataset and demo codes in the future. Please stay tuned. Thank you for your understanding.