Guibo Zhu

Collaborative Correlation Tracking

Correlation filter based tracking has attracted many researchers’ attention in recent years for high efficiency and robustness. Most existing works focus on exploiting different characteristics with correlation filters for visual tracking, e.g. circulant structure, kernel trick, effective feature representation and context information. However, how to handle the scale variation and the model drift is still an open problem. In this paper, we propose a collaborative correlation tracker to deal with the above problems. Firstly, we extend the correlation tracking filter by embedding the scale factor into the kernelized matrix to handle the scale variation. Then a novel long-term CUR filter for detection is learnt efficiently with random sampling to alleviate model drift by detecting effective object candidates in the collaborative tracker. In this way, the proposed approach could estimate the object state accurately and handle the model drift problem effectively. Extensive experiments show the superiority of the proposed method.

Weighted Part Context Learning for Visual Tracking

Context information is widely used in computer vision for tracking arbitrary objects. Most of the existing studies focus on how to distinguish the object of interest from background or how to use keypoint-based supporters as their auxiliary information to assist them in tracking. However, in most cases, how to discover and represent both the intrinsic properties inside the object and the surrounding context is still an open problem. In this paper, we propose a unified context learning framework that can effectively capture spatiotemporal relations, prior knowledge, and motion consistency to enhance trackers performance. The proposed weighted part context tracker (WPCT) consists of an appearance model, an internal relation model, and a context relation model. The appearance model represents the appearances of the object and the parts. The internal relation model utilizes the parts inside the object to directly describe the spatiotemporal structure property, while the context relation model takes advantage of the latent intersection between the object and background regions. Then, the three models are embedded in a max-margin structured learning framework. Furthermore, prior label distribution is added, which can effectively exploit the spatial prior knowledge for learning the classifier and inferring the object state in the tracking process. Meanwhile, we define online update functions to decide when to update WPCT, as well as how to reweight the parts. Extensive experiments and comparisons with the state of the arts demonstrate the effectiveness of the proposed method.

Person re-identification via rich color-gradient feature

Person re-identification refers to match the same pedestrian across disjoint views in non-overlapping camera networks. Lots of local and global features in the literature are put forward to solve the matching problem, where color feature is robust to viewpoint variance and gradient feature provides a rich representation robust to illumination change. However, how to effectively combine the color and gradient features is an open problem. In this paper, to effectively leverage the color-gradient property in multiple color spaces, we propose a novel Second Order Histogram feature (SOH) for person reidentification in large surveillance dataset. Firstly, we utilize discrete encoding to transform commonly used color space into Encoding Color Space (ECS), and calculate the statistical gradient features on each color channel. Then, a second order statistical distribution is calculated on each cell map with a spatial partition. In this way, the proposed SOH feature effectively leverages the statistical property of gradient and color as well as reduces the redundant information. Finally, a metric learned by KISSME [1] with Mahalanobis distance is used for person matching. Experimental results on three public datasets, VIPeR, CAVIAR and CUHK01, show the promise of the proposed approach.

Clustering based ensemble correlation tracking

A non-parametric sequential clustering is proposed for efficiently mining the low rank structure of historical objects represented by weighted cluster centers.To alleviate model drift, an adaptive object template is learned by the weighted clustered centers which can be used to calculate the spatial distribution of object and provide weakly supervised information for re-correcting the object state.A clustering based ensemble correlation tracker is proposed to jointly capture the target appearance by multi-scale kernelized correlation filter and exploit the long-term object properties by the object template with cluster analysis. Correlation filter based tracking has attracted many researchers attention in the recent years for its high efficiency and robustness. Most existing work has focused on exploiting different characteristics with correlation filter for visual tracking, e.g., circulant structure, kernel trick, effective feature representation and context information. Despite much success having been demonstrated, numerous issues remain to be addressed. Firstly, the target appearance model can not precisely represent the target in the tracking process because of the influence of scale variation. Secondly, online correlation tracking algorithms often encounter the model drift problem. In this paper, we propose a clustering based ensemble correlation tracker to deal with the above problems. Specifically, we extend the tracking correlation filter by embedding a scale factor into the kernelized matrix to handle the scale variation. Furthermore, a novel non-parametric sequential clustering method is proposed for efficiently mining the low rank structure of historical object representation through weighted cluster centers. Moreover, to alleviate the model drift, an object spatial distribution is obtained by matching the adaptive object template learned from the cluster centers. Similar to a coarse-to-fine search strategy, the spatial distribution is not only used for providing weakly supervised information, but also adopted to reduce the computational complexity in the detection procedure which can alleviate the model drift problem effectively. In this way, the proposed approach could estimate the object state accurately. Extensive experiments show the superiority of the proposed method.

Part Context Learning for Visual Tracking

Context information is widely used in computer vision for tracking arbitrary objects. Most existing works focus on how to distinguish the tracked object from background or inter-frame object similarity information or key-points supporters as their auxiliary information to assist them in tracking. However, in most cases, how to discover and represent both the intrinsic property inside the object and surrounding information is still an open problem. In this paper, we propose a unified context learning framework that can capture stable structure relations of in-object parts, context parts and the object itself to enhance the tracker’s performance. The proposed Part Context Tracker (PCT) consists of an appearance model, an internal relation model and an context relation model. The appearance model represents the appearances of the object and parts. The internal relation model utilizes the parts inside the object to describe the spatio-temporal structure property directly, while the context relation model takes advantage of the latent intersection between the object and background parts. Then the appearance model, internal relation model and context relation model are embedded in a max-margin structured learning framework. Furthermore, a simple robust update strategy using median filter is utilized, which can deal with appearance change effectively and alleviate the drift problem. Extensive experiments are conducted on various benchmark dataset, and the comparisons with state-of-the-arts demonstrate the effectiveness of our work.

Feature Distilled Tracking

Feature extraction and representation is one of the most important components for fast, accurate, and robust visual tracking. Very deep convolutional neural networks (CNNs) provide effective tools for feature extraction with good generalization ability. However, extracting features using very deep CNN models needs high performance hardware due to its large computation complexity, which prohibits its extensions in real-time applications. To alleviate this problem, we aim at obtaining small and fast-to-execute shallow models based on model compression for visual tracking. Specifically, we propose a small feature distilled network (FDN) for tracking by imitating the intermediate representations of a much deeper network. The FDN extracts rich visual features with higher speed than the original deeper network. To further speed-up, we introduce a shift-and-stitch method to reduce the arithmetic operations, while preserving the spatial resolution of the distilled feature maps unchanged. Finally, a scale adaptive discriminative correlation filter is learned on the distilled feature for visual tracking to handle scale variation of the target. Comprehensive experimental results on object tracking benchmark datasets show that the proposed approach achieves

Learning weighted part models for object tracking

{5\times }

Object tracking with part-based discriminative context models

speed-up with competitive performance to the state-of-the-art deep trackers.

Appearance features in Encoding Color Space for visual surveillance

Part models are formulated with a graph structure for object tracking.Weighted part models are used to handle target appearance change and occlusion.GMMs are used as weight models for describing dynamic evolution of object parts.Weight models are used to dynamically adjust the part appearance models.Weight models are used to control the sample selections for part model update. Despite significant improvements have been made for visual tracking in recent years, tracking arbitrary object is still a challenging problem. In this paper, we present a weighted part model tracker that can efficiently handle partial occlusion and appearance change. Firstly, the object appearance is modeled by a mixture of deformable part models with a graph structure. Secondly, through modeling the temporal evolution of each part with a mixture of Gaussian distribution, we present a temporal weighted model to dynamically adjust the importance of each part by measuring the fitness to the historical temporal distributions in the tracking process. Moreover, the temporal weighted models are used to control the sample selections for the update of part models, which makes different parts update differently due to partial occlusion or drastic appearance change. Finally, the weighted part models are solved by structural learning to locate the object. Experimental results show the superiority of the proposed approach.

Clustering Ensemble Tracking

Object tracking is a classic problem in computer vision. Part-based appearance model has been applied to object tracking and shown good performance. However, how to initialize the parts is still an open question. In this paper, we believe that the selection of discriminative parts and effectively modeling the structural context information could improve the tracking performance. Therefore, we tackle the tracking problem by discovering discriminative parts through exemplar-SVM in the initialization, and then exploit the structural relationship between discriminative context parts and the object in the process of tracking, which is consensual in the spatio-temporal domain. Experimental results demonstrate that our approach outperforms state-of-the-art trackers on benchmark videos.