In urban low-altitude environments, UAV visual odometry is hindered by dynamic interference, texture scarcity, and variable geometries, leading to unstable localization and reduced accuracy. To overcome these issues, we propose the Adaptive Geometry Fusion Ranging (AGF-VO) framework. AGF-VO extracts line features using an improved orthogonal representation of Plücker’s coordinates and integrates ORB point features to create a multimodal scene representation. An adaptive dynamic weighting mechanism adjusts the contributions of point and line features in real time based on scene complexity and matching quality. Additionally, the back-end optimization incorporates geometric consistency constraints—such as angles, distances, and polygon areas—to minimize joint reprojection error and enhance localization accuracy. Experimental results demonstrate that AGF-VO delivers high stability and robustness in environments with severe dynamic interference and sparse textures, thereby supporting autonomous UAV navigation and obstacle avoidance under GPS-denied conditions.