Multi-layer Rotation Memory Model-based correlation filter for visual tracking

Abstract

Object tracking technology is of great significance in laser image processing. However, occlusion or similar interference during visual object tracking may reduce the tracking precision or even cause tracking failure. Aiming at this issue, we propose a Multi-layer Rotation Memory Model-based Correlation Filter (MRMCF) for visual trackingin this paper. First, we establish a Multi-layer Rotation Memory (MRM) model, in which a set of three rotating concentric rings is used to simulate the three memory spacesand their updating processsimulate the memory spaces. Then we introduce the MRM model into the correlation filter tracking framework, which realizes realizing the dynamic updating of classifier parametersin the correlation filter. When the object is occluded or there is similar interference, the proposed tracker can use the Pre-occ classifier parameters stored in the memory spaces in the MRM model MRM memory spaces to retarget the object, thereby reducing the impact of these factors. The experimental results on the OTB50 dataset show that compared with trackers such as CNN-SVM, MEEM, Struck, etc., the proposed tracker achieves higher accuracy and success rate.