Unmanned aerial vehicle (UAV) cross-view object localization is widely used in applications such as field rescue, disaster detection, and express delivery. The feature-matching method, which locates the object position by matching the UAV object image and satellite image, is a key technology for UAV cross-view object localization. This study proposes a feature-matching method optimized using a rolling-guided filter. First, a rolling-guided filter is introduced to calculate the keypoint respondence, which addresses the problem of scale difference between the two images; then, we introduce the phase congruency information for feature description. By obtaining candidate corresponding keypoint sets through feature matching, the problem of gray level differences between images can be solved; Finally, we use the respondence of candidate keypoints to calculate the network loss, which addresses the problem of viewing angle differences between images. The experimental results demonstrate that the proposed method can effectively improve the feature-matching performance.