Deformable Target Tracking Algorithm based on Feature Complementary Learning

Aiming at the problem of a long loss of the tracking target of STAPLE (Sum of Template and Pixel-wise LEarners) algorithm, an ASTAPLE (Adjustable Staple) algorithm is proposed, which changes learning rate according to the Peak Sidelobe Ratio of the image response diagram. In order to compare proposed tracking algorithm with other correlation filtering tracking algorithms, such as KCF, DSST, STAPLE, the experiment is made on OTBIOO benchmark for tracking deformable target. The precision of ASTAPLE tracking algorithm is higher than STAPLE tracking algorithm’s and the success rate of ASTAPLE tracking algorithm higher than STAPLE tracking algorithm’s. Moreover, the success rate of ASTAPLE tracking algorithm is higher than KCF tracking algorithm’s and DSST tracking algorithm’s. Therefore, the proposed tracking algorithm is able to track deformable target more effectively in the long-term tracking with higher robustness.