Dynamic Foreground Perception Correlation Filter with Spatiotemporal Regularization Term

This is the authors' abstract. We don't add key points for this paper.

In response to the problem of poor tracking accuracy and susceptibility to temporal degradation in the presence of similar background interference in existing correlated filtering object tracking algorithms, a dynamic foreground perception correlation filter with spatiotemporal regularization term is proposed. Firstly, incorporating a temporal regularizer on the basis of DRCF to alleviate the problem of model temporal degradation; Secondly, a anomaly perception function is proposed, which uses a Kalman filter to correct the current frame tracking result when the correlation filter tracking result is unreliable. Finally, comparative experiments were conducted on the OTB-50 and OTB-100 datasets with various representative correlation filtering tracking algorithms. The experimental results showed that the algorithm proposed in this paper has good tracking performance, achieving a tracking speed of 49.18 fps.