Abstract:In scenes high in visual complexity, the identification of a moving object can be affected by changes in scale and occlusion factors during the tracking process, resulting in reduced tracking accuracy. Accordingly, to address the problem of low accuracy, a multiscale fast correlation filtering tracking algorithm based on a feature fusion model is proposed in the present work with the aim of reducing the poor tracking effects caused by occlusion discrimination and scale changes in complex scenes. The object's grayscale (GRAY) feature, histogram of oriented gradient (HOG) feature, and color name (CN) feature are reduced to dimensions and fused to form a feature matrix. Moreover, a hierarchical principal component analysis (HPCA) algorithm is used to extract visually salient features and reconstruct the feature matrix under real-time conditions, the correlation filtering position is trained, the number of dimensions is effectively reduced, and the feature fusion matrix is used to train the multiscale fast correlation filtering, with the result that the object's position and scale can be accurately predicted. The proposed algorithm is then compared with five popular correlation filtering tracking algorithms. Experimental results demonstrate that its average tracking speed reaches a reasonable frames/second velocity; moreover, it can also achieve promising object tracking results on the OTB benchmark data sets. The tracking accuracy is superior to that of the other five correlation filtering tracking algorithms when applied to scenes featuring object occlusion and changes in scale. The proposed algorithm also exhibits better robustness and improved performance under real-time conditions in sophisticated scenarios, including scale variation, deformation, fast motion, occlusion, and so on.

Retracted: Fast Compressive Tracking With Robust Example Selection Based On Multiple Instance Learning In Smart And Autonomous Systems (Retracted Article. See Vol. 76, Pg. 761, 2018)

WITHDRAWN: Fast Compressive Tracking with Robust Example Selection Based on Multiple Instance Learning in Smart and Autonomous Systems

RETRACTED: Multiscale fast correlation filtering tracking algorithm based on a feature fusion model (Retracted article. See vol. 34, 2022)

Explicitly Modeling the Discriminability for Instance-Aware Visual Object Tracking

Visual Tracking Via Online Discriminative Multiple Instance Metric Learning

Visual Object Tracking Based on Multi-exemplar Regression Model

Robust Visual Tracking Using Dynamic Classifier Selection with Sparse Representation of Label Noise

Multiple Instance Models Regression for Robust Visual Tracking

Online Discriminative Tracking with Active Example Selection

Robust Visual Tracking Via Randomly Projected Instance Learning

Adaptive Resolution Optimization and Tracklet Reliability Assessment for Efficient Multi-Object Tracking

Robust Visual Tracking Via Multiple Discriminative Models with Object Proposals

Object Tracking of Online Feature Selection Multiple Instance Learning Via Minimizing Trace

Efficient Compressive Sensing Tracking Via Mixed Classifier Decision

Coupling Semi-Supervised Learning And Example Selection For Online Object Tracking

Multiple Instance Learning Tracking Method with Local Sparse Representation

Online Filtering Training Samples for Robust Visual Tracking.

Real-time Object Tracking Via Compressive Feature Selection

Object Tracking Via Online Multiple Instance Learning with Reliable Components

Visual Tracking with Online Structural Similarity-Based Weighted Multiple Instance Learning

Learning a Reliable Decision Making Policy for Robust Tracking