From self-driving cars to virtual reality, modern technologies rely on precise localization and environmental perception. Panoramic cameras are ideal tools thanks to their ultra-wide field of view, but the trade-off comes in the form of geometric distortions. Traditional projection models—such as equidistant cylindrical or cube mapping—fail to preserve spatial consistency, causing errors in feature detection and motion estimation. For visual-inertial odometry (VIO) systems, which integrate camera and inertial data to estimate movement, such distortions can be detrimental. The inability to maintain accurate geometric relationships across the visual field undermines tracking reliability and limits deployment in complex scenarios. Due to these challenges, there is a growing need to redesign projection models with spatial integrity at their core.
In a study (DOI: 10.1186/s43020-025-00171-y) published in Satellite Navigation (July 2025), researchers from Sun Yat-sen University introduced Geotri-VIO, a next-generation panoramic VIO framework. The team developed a novel multi-prism projection approach that restores geometric consistency across panoramic images. By shaping projection planes as tangent triangular prisms around the camera’s spherical field, the system ensures accurate spatial mapping and feature tracking. Validated against public and real-world datasets, Geotri-VIO demonstrated superior accuracy and robustness compared to leading systems like LF-VIO and VINS-Mono, marking a breakthrough in wide-angle visual localization.
At the heart of Geotri-VIO is a mathematically grounded innovation: a triangular prism projection model that aligns each image plane tangentially with the panoramic camera’s sphere. This design preserves spatial relationships between pixels, correcting the nonlinear distortions common in omnidirectional projections. The researchers evaluated prism models with different face counts and found the triangular configuration achieved the best balance between local and global geometric consistency.
Extensive experiments showed that this projection greatly improves the performance of point and line feature extraction—key elements for pose estimation. Compared to other models, Geotri-VIO demonstrated higher tracked feature ratios and reduced pose and trajectory errors. For example, on both the indoor PALVIO dataset and outdoor real-world sequences, the system achieved up to 39% lower Absolute Trajectory Error and superior robustness in dynamic conditions.
Notably, Geotri-VIO is designed to integrate both point-based and point-line-based tracking methods, making it highly versatile. The system also runs efficiently, with negligible computational overhead from the projection process. This allows it to operate in real time, even under the demands of mobile robotics and autonomous driving. Its performance advantage stems from one key insight: respecting the true geometry of panoramic space.
“Our approach was to rethink how panoramic images should be interpreted—by embracing geometry, not distorting it,” said Dr. Hui Cheng, corresponding author of the study. “The triangular prism projection restores the spatial fidelity that traditional models lose. As a result, systems using Geotri-VIO can navigate more precisely, even in complex or unpredictable environments. We believe this framework will serve as a foundation for the next generation of vision-based localization technologies.”
Geotri-VIO’s robust, distortion-free vision system has broad potential across industries. In robotics and autonomous driving, it provides more reliable localization in urban, indoor, or GPS-blocked environments. For augmented and virtual reality, the improved spatial consistency ensures more stable and immersive experiences. The framework’s flexibility—compatible with both point and line features—means it can scale across different platforms and tasks. Crucially, the low computational load makes it suitable for real-time, embedded applications. As panoramic imaging continues to expand in use, Geotri-VIO offers a timely and transformative solution for accurate, wide-field visual perception.
###
References
DOI
10.1186/s43020-025-00171-y
Original Source URL
https://doi.org/10.1186/s43020-025-00171-y
Funding Information
This work was supported in part by the China National Key Research and Development Program under Grant 2022YFB3903804.
About Satellite Navigation
Satellite Navigation (E-ISSN: 2662-1363; ISSN: 2662-9291) is the official journal of Aerospace Information Research Institute, Chinese Academy of Sciences. The journal aims to report innovative ideas, new results or progress on the theoretical techniques and applications of satellite navigation. The journal welcomes original articles, reviews and commentaries.