High-precision Sub-pixel Object Tracking Algorithm

Aiming at the problem that the positioning accuracy of the KCF tracking algorithm is difficult to reach the pixel level and cannot adapt to the target scale variation well, the correlation filter tracking algorithm based on the scale pyramid achieves higher tracking accuracy, but the tracking speed is greatly reduced. Introducing the logarithmic polar coordinate transformation of the image, an object tracking algorithm based on the logarithmic polar coordinate transformation is proposed. First, the target template is transformed into the logarithmic polar coordinate, and the scale variation of the target is converted into a displacement signal, then extract the HOG features before and after the target template transformation, and the filter model of displacement and scale is established. Finally, the displacement and scale factor of the object are tracked synchronously under the framework of correlation filtering, and the two are merged to obtain the target tracking frame. The experimental results show that: The average overlap precision of the algorithm in this paper is high, and the tracking effect is better (Experiment 1). The algorithm in this paper can track rigid objects stably and can adapt to scale variation accurately very well (Experiment 2). The overall accuracy and success rate are in the first place (Experiment 3). The algorithm in this paper can approximately achieve pixel-level positioning accuracy, and the tracking speed can reach twice the traditional algorithm (Experiment 4).