In recent years, leader-follower Unmanned Aerial Vehicle (UAV) swarms have seen increasingly widespread application across various domains. However, when follower UAVs fly at low altitudes in environments such as dense urban areas or canyons, the Global Navigation Satellite System (GNSS) signals are often severely interfered with or blocked, leading to a decrease in localization accuracy or even complete failure. To address this problem, this paper proposes an innovative visual cooperative navigation framework. It leverages the multi-scale perception capabilities of a leader UAV (global view) and a follower UAV (local view) to achieve cross-view visual cooperative localization. Specifically, the proposed method extracts and matches Speeded Up Robust Features (SURF) from the downward-facing images captured by the onboard cameras of the leader and follower UAVs to establish a mapping relationship between the images. Subsequently, it utilizes the leader’s positioning information to locate the follower. To address the challenges of field of view (FOV) disparities and ineffective matching regions between the leader and follower, this work presents a SURF feature matching method based on a Dynamic Region of Interest (DRoI). It crops the leader’s image using the DRoI, thereby processing only the overlapping FOV of the two UAVs, which significantly reduces computational complexity and processing time. Experimental results demonstrate that the proposed visual cooperative navigation framework exhibits excellent localization accuracy and real-time performance, effectively addressing the localization challenges for follower UAVs in GNSS-interfered environments.