Abstract: To address the challenges in multi-object tracking of drone swarms under complex environments—including high appearance similarity, frequent mutual occlusions, and the current research focus being largely limited to single-drone tracking or direct adaptation of generic multi-object tracking algorithms—this paper presents UAVSwarmV2, a large-scale enhanced dataset. Built upon the original UAVSwarm, UAVSwarmV2 expands the image data by 70%, covers diverse dynamic scenarios, and features significantly improved resolution. We also propose a dual matching strategy that integrates topology-aware and geometric information. By combining an improved CIoU-based geometric similarity with a local neighborhood topological similarity, the method effectively mitigates association ambiguity in densely packed drone formations. Experiments show that, compared to the baseline ByteTrack, our approach improves the Identity F1 score (IDF1) by 5.7% and reduces identity switches (ID Switch) from 349 to 96, demonstrating its effectiveness. The dataset will be publicly released at: https://github.com/UAVSwarm/UAVSwarmV2-dataset.git.