Class
Article
College
College of Engineering
Department
Computer Science Department
Presentation Type
Oral Presentation
Abstract
Sparse representation has recently been successfully applied in visual tracking. It utilizes a set of templates to represent target candidates and find the best one with the minimum reconstruction error as the tracking result. In this presentation, we propose a robust deep features-based structured group local sparse tracker, which exploits the convolutional neural network (CNN) features of local patches inside target candidates and represents them by a set of templates in the particle filter framework. To extract the local CNN features, we first set the size of each target candidate to 64 by 64 pixels to contain sufficient object-level information with decent resolution. Then, we pass each target candidate to the pre-trained VGG19 network to automatically extract their representative global features. Finally, we employ the concept of shared features to divide the global feature map into pre-defined number of overlapping local features maps. Unlike the conventional local sparse trackers, the proposed optimization model in our tracker employs a group-sparsity regularization term to seamlessly adopt local and spatial information of the target candidates and attain the spatial layout structure among them. To solve the optimization model, we propose an efficient and fast numerical algorithm that consists of two subproblems with the closed-form solutions using alternating direction method of multipliers (ADMM). Different evaluations in terms of success and precision on the benchmarks of challenging image sequences (e.g., OTB50 and OTB100) demonstrate the superior performance of the proposed tracker against several state-of-the-art trackers.
Location
Room 154
Start Date
4-10-2019 1:30 PM
End Date
4-10-2019 2:45 PM
Included in
Robust Structured Group Local Sparse Tracker Using Convolutional Neural Network Features
Room 154
Sparse representation has recently been successfully applied in visual tracking. It utilizes a set of templates to represent target candidates and find the best one with the minimum reconstruction error as the tracking result. In this presentation, we propose a robust deep features-based structured group local sparse tracker, which exploits the convolutional neural network (CNN) features of local patches inside target candidates and represents them by a set of templates in the particle filter framework. To extract the local CNN features, we first set the size of each target candidate to 64 by 64 pixels to contain sufficient object-level information with decent resolution. Then, we pass each target candidate to the pre-trained VGG19 network to automatically extract their representative global features. Finally, we employ the concept of shared features to divide the global feature map into pre-defined number of overlapping local features maps. Unlike the conventional local sparse trackers, the proposed optimization model in our tracker employs a group-sparsity regularization term to seamlessly adopt local and spatial information of the target candidates and attain the spatial layout structure among them. To solve the optimization model, we propose an efficient and fast numerical algorithm that consists of two subproblems with the closed-form solutions using alternating direction method of multipliers (ADMM). Different evaluations in terms of success and precision on the benchmarks of challenging image sequences (e.g., OTB50 and OTB100) demonstrate the superior performance of the proposed tracker against several state-of-the-art trackers.