Robust visual tracking via multi-feature response maps fusion using a collaborative local-global layer visual model

Robust Visual Tracking Using Structural Patch Response Map Fusion Based on Complementary Correlation Filter and Color Histogram

“…Lukezic et al [28] construct a spatial reliability map to adjust the filter support to the part of the target object suitable for tracking by exploiting color histograms. Zhang et al [15] propose a collaborative local-global layer visual tracking method (LGCmF), in which a block tracker (SLC) utilizing structural local color histograms feature and a global correlation filter tracker based on HOG feature are merged in the response map level. Inspired by [15], the block strategy also is adopted in this work.…”

“…Zhang et al [15] propose a collaborative local-global layer visual tracking method (LGCmF), in which a block tracker (SLC) utilizing structural local color histograms feature and a global correlation filter tracker based on HOG feature are merged in the response map level. Inspired by [15], the block strategy also is adopted in this work. In contrast to [15] that only applies part-based tracking strategy in color histogram model, we employ more complete blocking strategy in both component trackers and more efficient block weighting method for each patch based on adaptive hedge algorithm.…”

“…Inspired by [15], the block strategy also is adopted in this work. In contrast to [15] that only applies part-based tracking strategy in color histogram model, we employ more complete blocking strategy in both component trackers and more efficient block weighting method for each patch based on adaptive hedge algorithm.…”

“…These tracking failures also lead to the failure of Staple tracker since it has no emphasis on reliable component tracker. The LGCmF tracker [15], which is an improved method based on the Staple, performs well on both Surfer and Box sequences, while fails on the BlurCar1 and Shaking sequences.…”

“…Then based on the tracking reliabilities of CFSP and CHSP, we selectively combine the response maps of them to formulate the final complementary structural patches response fusion tracker (CSPRF). Inspired by [15,17], we train and update a SVM detector for determining the confidences of the component trackers CFSP and CHSP, and implement a re-detect procedure when both of the component trackers are unreliable. From Figure 1, it can be seen that our proposed CSPRF tracker performs favorably when the Staple and LGCmF lose the target.…”