Abstract
In this paper, we propose a novel tracking method based on structured metric learning, which takes the advantages of both structured learning and distance metric learning. In our method, tracking is formulated as a structured metric learning problem, which not only considers the importance of different samples, but also improves the discriminability by learning a specific distance metric for matching. Specifically, a concrete structured metric learning method is realized by making use of the constraints from the target and its neighbour training samples under the above framework. Besides, a closed-form solution is derived for the structured metric learning problem. To improve the matching robustness, the K-nearest neighbours (KNN) distance is employed to determine the final tracking result. Experimental results in the benchmark dataset demonstrate that the proposed structured metric learning based tracking method can achieve desirable performance.
Highlights
Object tracking is one of the research topics in computer vision, multimedia information processing, etc
With the selected samples {xm}, we introduce the K-nearest neighbours (KNN) distance to calculate the distances between the candidate samples and the target subspace and determine the final tracking result
Both the training samples and the candidate samples are normalized to the fixed size 30 × 30, which is set as the same size as in many other tracking methods [4], [9] and accords with the size of histogram of orientation gradients (HOG)
Summary
Object tracking is one of the research topics in computer vision, multimedia information processing, etc. Zhang et al [34] introduce fuzzy learning into tracking and propose a fuzzy least squares SVM for tracking, which can effectively deal with the fuzzy boundary problem between the training samples These discriminative methods address the issues in the traditional methods by different manners, the metrics hiding in these methods are still defined in advance or fixed, which lacks of the flexibility and accuracy to represent the appearances of different targets and scenes. We propose a novel tracking method based on structured metric learning, which considers the importance of different samples, and improves discriminability by learning a specific distance metric for matching. It can be seen that M can be obtained in closed form, which can be calculated efficiently
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
