Search

In this paper we present a novel visual tracking algorithm, in which object tracking is achieved by using subspace learning and Huber loss regularization in a particle filter framework. The changing appearance of tracked target is modeled by principle component analysis basis vectors and row group sparsity. This method takes advantage of the strengths of subspace representation and explicitly takes the underlying relationship between particle candidates into consideration in the tracker. The representation of each particle is learned via the multi-task sparse learning method. Huber loss function is employed to model the error between candidates and templates, yielding robust tracking. We utilize the alternating direction method of multipliers to solve the proposed representation model. In experiments we tested sixty representative video sequences that reflect the specific challenges of tracking and used both qualitative and quantitative metrics to evaluate the performance of our tracker. The experiment results demonstrated that the proposed tracking algorithm achieves superior performance compared to nine state-of-the-art tracking methods. [ABSTRACT FROM AUTHOR]

This paper proposes an online object tracking algorithm in which the object tracking is achieved by using multi-task sparse learning and non-negative matrix factorization under the particle filtering framework. The object appearance is first modeled by subspace learning to reflect the target variations across frames. Combination of non-negative components is learned from examples observed in previous frames. In order to robust tracking an object, group sparsity constraints are included to the non-negativity one. Furthermore, the alternating direction method of multipliers algorithm is employed to compute the model efficiently. Qualitative and quantitative experiments on a variety of challenging sequences show favorable performance of the proposed algorithm against state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Abstract Existing convolutional neural network (CNN) based trackers have limited tracking performance in some challenging scenarios such as deformation, background clutter and illumination variation, because the features extracted from a single layer or from a linear combination of multiple layers are insufficient to describe the target appearance. To overcome this problem, we propose a novel tracking algorithm based on interactive multiple model (IMM) framework for better exploring deep features from different layers (IMM_DFT). In this method, we first build measurement models from convolutional layers by applying correlation filters on hierarchical features. Then, to effectively estimate the target state for each layer, we design a hybrid system which consists of the foregoing measurement model and an online learning motion model. Finally, in order to achieve the optimal fusion of the systems for adapting diverse appearance variation of the target and background, an IMM estimator is developed to dynamically adjust the weight of each system using likelihood function and transition probabilities. Extensive experiments on OTB-2013, OTB-2015 and VOT-2016 benchmark databases demonstrate that the proposed algorithm achieves more favorable performance than several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• A novel uncertain motion tracking framework with semantics estimation and region proposals is proposed. • A semantics object proposals generation strategy is proposed. • The proposed hybrid semantics tracking algorithm combines the full advantages of globally sparse semantics region proposals prediction and correlation filter prediction. • The proposed semantics-contextual estimation method can generate more credible candidates. • Experimental results show the state-of-the-art performance of our method. In real world, most of the tracking methods suffer from uncertain motion, which may make the tracker failure because of a local search window with the motion smooth assumption. To address this problem, a novel tracking framework based on convolutional net with semantics estimation and region proposals is proposed. Firstly, we present a semantics object proposals generation strategy, including category-level semantics proposals, one-object-level semantics estimation and semantics-contextual proposals generation, to obtain a few of high-quality object-oriented proposals covering uncertain motion. Secondly, combining the globally sparse semantics region proposals prediction and correlation filter prediction, a hybrid semantics tracking algorithm is proposed, which obtains a coarse object location by the decision of multiple response maps. Finally, we learn and train independent correlation filter to estimate the scale of target for a higher tracking accuracy. Extensive experiments on two visual tracking benchmarks and results demonstrate our method achieves state-of-the-art performance. [ABSTRACT FROM AUTHOR]