Visual navigation based on prior structured objects: monocular, IMU-aided, and stereo fusion strategies

To address the vulnerability of GNSS in complex environments, this paper proposes a visual navigation framework based on prior structured objects. The framework includes monocular navigation using sparse repetitive features, IMU-aided monocular navigation, and stereo navigation based on discrete interval features, targeting different scene scales and complexities. Experimental results show that the proposed monocular method achieves a position RMSE within 1.90 mm and a heading error below 0.11°. The IMU-aided method further enhances robustness, with a maximum position error of 2.64 mm. The stereo method achieves positioning errors below 1 m in all spatial directions and supports processing speeds up to 38 fps, demonstrating high accuracy and real-time performance.