Anti-noise Object Tracking Based on Siamese Octave Convolution and Attentional Correlation Fusion

The task of object tracking is to calculate the location of the selected object in the video sequence, but it will face the problem of performance reduction in the noise environment. The goal of this paper is to enhance the noise resistance of the tracking network without increasing the computational burden. The Siamese Octave convolution module is proposed to improve the noise resistance of CNN features and to further enhance the self-similarity of template feature map and search feature map. At the same time, a new cross-correlation fusion method is introduced, which improves the multi-scale of cross-correlation results by aggregating local and global context information, and brings better adaptability at various noise levels. Through the use of our method on the anchor-free tracker Ocean, the experimental results show that the improved method proposed in this paper can effectively improves the noise resistance of the object tracking model.