Towards long-term UAV object tracking via effective feature matching

Object tracking based on unmanned aerial vehicles (UAVs) has attracted extensive research attention recently since it provides the ability to continuously observing and tracking the target owing to its inherent advantage. However, occlusion is a crucial interference which may cause performance degradation in long-term UAV-based tracking. In this Letter, the authors propose a robust and efficient long-term tracker based upon local feature matching and density clustering. To be more specific, the authors propose a key-point-matching based confidence indicator to monitor the tracking condition and activate the re-detection module when occlusion is predicted. Once occlusion occurs, a novel density-based clustering method is utilised to re-locate the target with the collected local features. Extensive experiments have demonstrated that the proposed algorithm performs favourably against the other related trackers.