Small Moving Object Tracking in Dynamic Video

The problems of moving object detecting and tracking on UAV (Unmanned Aerial Vehicle) platforms in surveillance operations are addressed, and a robust tracking method for reliably detecting and tracking of foreground moving objects in a dynamically changing environment is proposed. Firstly, using the modified robust local feature detector SURF (Speeded Up Robust Features) which is linear ordering defined on the set of correspondences, the PROSAC (Progressive Sample Consensus) algorithm exploits by a similarity function to match the detected key points in consecutive frames and establishes correspondences. Secondly, target tracking is realized by matching the invariant feature points and the scale of object is characterized by scale information of the corresponding points. Finally, a Kalman filter tracker is adopted. Experiments on challenging sequences show the good performance of the proposed method and it is capable of tracking targets as small as 30 pixels.