A novel foreground region analysis using NCP-DBP texture pattern for robust visual tracking

Robust visual tracking is the important stage in the computer vision applications such as robotics, man-free control systems, and the visual surveillance. Accurate motion states estimation and the target representation in visual tracking system are based on the appearances of the target. The factor affects the learning of target representation is the accumulated error due to the pose, illumination changes, and the uneven background. The presence of dynamic background and the shadowing effects causes the visual drift and destructive information. Besides, the misclassification of target region induces the false detection of moving objects. The K-means and Fuzzy-C-means clustering algorithms are available to segment the foreground/background and suppress the shadow region on the basis of the non-changing background of the surveillance area. This paper proposes the novel background normalization technique with textural pattern analysis to suppress the shadow region. The Neighborhood Chain Prediction (NCP) algorithm is used to cluster the uneven background and the Differential Boundary Pattern (DBP) extracts the texture of the video frame to suppress the shadow pixels present in the frame. The lower intensity estimation and the prediction of the area around the lower intensity in proposed work enhance the pixels for shadow removal. The shadow-free frame split up into several grids and the histograms of features are extracted from the grid formatted frame. Finally, the Machine Level Classification (MLC) finds the matching grid corresponds to the tracking region and provides the binary labeling to separate the background and foreground. The proposed DBP-based visual tracking system is high robustness over the sudden illumination changes and the dynamic background due to the texture pattern analysis. The comparison of proposed NCP-DBP combination with the existing segmentation techniques regarding the accuracy, precision, recall, F-measure, success and error rate assured the effectiveness in visual tracking applications.     

Multimodal MSEPF for visual tracking

Recently, particle filter has been applied to many visual tracking problems and it has been modified in order to reduce the computation time or memory usage. One of them is the Mean-Shift embedded particle filter (MSEPF, for short) and it is further modified as Randomized MSEPF. These methods can decrease the number of the particles without the loss of tracking accuracy. However, the accuracy may depend on the definition of the likelihood function (observation model) and of the prediction model. In this paper, the authors propose an extension of these models in order to increase the tracking accuracy. Furthermore, the expansion resetting method, which was proposed for mobile robot localization, and the changing the size of the window in Mean-Shift search are also selectively applied in order to treat the occlusion or rapid change of the movement.     

Active matching for visual tracking

In the feature matching tasks which form an integral part of visual tracking or SLAM (Simultaneous Localisation And Mapping), there are invariably priors available on the absolute and/or relative image locations of features of interest. Usually, these priors are used post-hoc in the process of resolving feature matches and obtaining final scene estimates, via ‘first get candidate matches, then resolve’ consensus algorithms such as RANSAC or JCBB. In this paper we show that the dramatically different approach of using priors dynamically to guide a feature by feature matching search can achieve global matching with far fewer image processing operations and lower overall computational cost. Essentially, we put image processing into the loop of the search for global consensus. In particular, our approach is able to cope with significant image ambiguity thanks to a dynamic mixture of Gaussians treatment. In our fully Bayesian algorithm denoted Active Matching, the choice of the most efficient search action at each step is guided intuitively and rigorously by expected Shannon information gain. We demonstrate the algorithm in feature matching as part of a sequential SLAM system for 3D camera tracking with a range of settings, and give a detailed analysis of performance which leads to performance-enhancing approximations to the full algorithm.     

Neural structures for visual motion tracking

This paper addresses visual motion tracking by a connectionist method, and aims at showing how the flexibility and the generalization power of neural networks can enhance a tracking system's adaptiveness and effectiveness. The simple principle of operation widens the range of applicability. A set of tracking structures that exhibit increasing levels of integration and efficiency are described. We also show how multinetwork architectures for estimate averaging may greatly increase tracking stability. The validity of the basic mechanism was assessed on a simple domain; however, a specific difficult testbed made it possible to verify the effectiveness of the method.     

Collaborative strategy for visual object tracking

Multimedia Tools and Applications - Adaptively learning the difference between object and background, discriminative trackers are able to overcome the complex background problem in visual object...     

Dynamic active contours for visual tracking

Visual tracking using active contours is usually set in a static framework. The active contour tracks the object of interest in a given frame of an image sequence. A subsequent prediction step ensures good initial placement for the next frame. This approach is unnatural; the curve evolution gets decoupled from the actual dynamics of the objects to be tracked. True dynamical approaches exist, all being marker particle based and thus prone to the shortcomings of such particle-based implementations. In particular, topological changes are not handled naturally in this framework. The now classical level set approach is tailored for evolutions of manifolds of codimension one. However, dynamic curve evolution is at least a codimension two problem. We propose an efficient, level set based approach for dynamic curve evolution, which addresses the artificial separation of segmentation and prediction while retaining all the desirable properties of the level set formulation. It is based on a new energy minimization functional which, for the first time, puts dynamics into the geodesic active contour framework.     

Improved assignment with ant colony optimization for multi-target tracking

Detecting and tracking ground targets is crucial in military intelligence in battlefield surveillance. Once targets have been detected, the system used can proceed to track them where tracking can be done using Ground Moving Target Indicator (GMTI) type indicators that can observe objects moving in the area of interest. However, when targets move close to each other in formation as a convoy, then the problem of assigning measurements to targets has to be addressed first, as it is an important step in target tracking. With the increasing computational power, it became possible to use more complex association logic in tracking algorithms. Although its optimal solution can be proved to be an NP hard problem, the multidimensional assignment enjoyed a renewed interest mostly due to Lagrangian relaxation approaches to its solution. Recently, it has been reported that randomized heuristic approaches surpassed the performance of Lagrangian relaxation algorithm especially in dense problems. In this paper, impelled from the success of randomized heuristic methods, we investigate a different stochastic approach, namely, the biologically inspired ant colony optimization to solve the NP hard multidimensional assignment problem for tracking multiple ground targets.     

Online unsupervised feature learning for visual tracking

We propose a method for visual tracking-by-detection based on online feature learning. Our learning framework performs feature encoding with respect to an over-complete dictionary, followed by spatial pyramid pooling. We then learn a linear classifier based on the resulting feature encoding. Unlike previous work, we learn the dictionary online and update it to help capture the appearance of the tracked target as well as the background. In more detail, given a test image window, we extract local image patches from it and each local patch is encoded with respect to the dictionary. The encoded features are then pooled over a spatial pyramid to form an aggregated feature vector. Finally, a simple linear classifier is trained on these features.Our experiments show that the proposed powerful—albeit simple—tracker, outperforms all the state-of-the-art tracking methods that we have tested. Moreover, we evaluate the performance of different dictionary learning and feature encoding methods in the proposed tracking framework, and analyze the impact of each component in the tracking scenario. In particular, we show that a small dictionary, learned and updated online is as effective and more efficient than a huge dictionary learned offline. We further demonstrate the flexibility of feature learning by showing how it can be used within a structured learning tracking framework. The outcome is one of the best trackers reported to date, which facilitates the advantages of both feature learning and structured output prediction. We also implement a multi-object tracker, which achieves state-of-the-art performance.     

多相关滤波自适应融合的鲁棒目标跟踪

目的 由于目标在复杂场景中可能会发生姿态变化、物体遮挡、背景干扰等情况,目标跟踪仍然是一个具有挑战性的课题。目前判别性相关滤波方法在目标跟踪问题上获得了成功而又广泛的应用。标准的相关滤波方法基于循环偏移得到大量训练样本,并利用快速傅里叶变换加速求解滤波器,使其具有很好的实时性和鲁棒性,但边界偏移带来的消极的训练样本降低了跟踪效果。空间正则化的相关滤波跟踪方法引入空间权重函数,增强目标区域的滤波器作用,在增大了目标搜索区域的同时,也增加了计算时间,而且对于目标形变不规则,背景相似的情景也会增强背景滤波器,从而导致跟踪失败。为此,基于以上问题,提出一种自适应融合多种相关滤波器的方法。方法 利用交替方向乘子法将无约束的相关滤波问题转化为有约束问题的两个子问题,在子问题中分别采用不同的相关滤波方法进行求解。首先用标准的相关滤波方法进行目标粗定位,进而用空间正则化的相关滤波跟踪方法进行再定位,实现了目标位置和滤波模板的微调,提高了跟踪效果。结果 本文算法和目前主流的一些跟踪方法在OTB-2015数据集中100个视频上,以中心坐标误差和目标框的重叠率为评判标准进行了对比实验,本文算法能较好地处理多尺度变化、姿态变化、背景干扰等问题,在CarScale、Freeman4、Girl等视频上都表现出了最好的跟踪结果;本文算法在100个视频上的平均中心坐标误差为28.55像素,平均目标框重叠率为61%,和使用人工特征的方法相比,均高于其他算法,与使用深度特征的相关滤波方法相比,平均中心坐标误差高了6像素,但平均目标框的重叠率高了4%。结论 大量的实验结果表明,在目标发生姿态变化、尺度变化等外观变化时,本文算法均具有较好的准确性和鲁棒性。     

Learning structured visual dictionary for object tracking

In this paper, we propose a visual tracking algorithm by incorporating the appearance information gathered from two collaborative feature sets and exploiting its geometric structures. A structured visual dictionary (SVD) can be learned from both appearance and geometric structure, thereby enhancing its discriminative strength between the foreground object and the background. Experimental results show that the proposed tracking algorithm using SVD (SVDTrack) performs favorably against the state-of-the-art methods.     

Probabilistic fusion-based parameter estimation for visual tracking

In object tracking, visual features may not be discriminative enough to estimate high dimensional motion parameters accurately, and complex motion estimation is computationally expensive due to a large search space. To tackle these problems, a reasonable strategy is to track small components within the target independently in lower dimensional motion parameter spaces (e.g., translation only) and then estimate the overall high dimensional motion (e.g., translation, scale and rotation) by statistically integrating the individual tracking results. Although tracking each component in a lower dimensional space is more reliable and faster, it is not trivial to combine the local motion information and estimate global parameters in a robust way because the individual component motions are frequently inconsistent. We propose a robust fusion algorithm to estimate the complex motion parameters using variable-bandwidth mean-shift. By employing correlation-based uncertainty modeling and fusion of individual components, the motion parameter that is robust to outliers can be detected with variable-bandwidth density-based fusion (VBDF) algorithm. In addition, we describe a method to update target appearance model for each component adaptively based on the component motion consistency. We present various tracking results and compare the performance of our algorithm with others using real video sequences.     

特征融合与视觉目标跟踪 *

针对跟踪过程中各类图像特征分离背景和目标能力的变化 ,提出一种基于增量判别分析的特征融合算法。该算法首先计算各特征图像的似然图 ,然后通过增量判别分析计算各特征分类性能 ,得到相应权重 ,并在此基础上求取融合似然图 ,通过粒子滤波算法确定待跟踪目标状态。通过对可见光及红外成像视频序列的仿真表明,该算法对环境光照变化、视角变化以及局部遮挡等均具有一定的鲁棒性。     

A dual-kernel-based tracking approach for visual target

A dual-kernel-based tracking approach for visual target is proposed in this paper. The similarity between candidate and target model, and the contrast between candidate and its neighboring background are considered simultaneously when evaluating a target candidate. The similarity is measured by Bhattacharyya coefficient while the contrast is calculated with Jensen-Shannon divergence, and they are adaptively fused into a novel objective function. By maximizing the linear approximation of objective function, a dual-kernel target location-shift relation from current location to a new location is induced. According to the location-shift relation, the optimal target location can be recursively gained in the mean shift procedure. Experimental evaluations on several image sequences demonstrate that the proposed algorithm can gain more accurate target location and better identification power to false target, and it is also robust to deformation and partial occlusion.     

基于多层特征增强的实时视觉跟踪

为了解决全卷积孪生视觉跟踪网络（SiamFC）出现相似语义信息干扰物使得跟踪目标发生漂移,导致跟踪失败的问题,设计出一种基于多层特征增强的实时视觉跟踪网络（MFESiam）,分别去增强高层和浅层的特征表示能力,从而提升算法的鲁棒性。首先,对于浅层特征,利用一个轻量并且有效的特征融合策略,通过一种数据增强技术模拟一些在复杂场景中的变化,例如遮挡、相似物干扰、快速运动等来增强浅层特征的纹理特性;其次,对于高层特征,提出一个像素感知的全局上下文注意力机制模块（PCAM）来提高目标的长时定位能力;最后,在三个具有挑战性的跟踪基准库OTB2015、GOT-10K和2018年视觉目标跟踪库（VOT2018）上进行大量实验。实验结果表明,所提算法在OTB2015和GOT-10K上的成功率指标比基准SiamFC分别高出6.3个百分点和4.1个百分点,并且以每秒45帧的速度运行达到实时跟踪。在VOT2018实时挑战上,所提算法的平均期望重叠率指标超过2018年的冠军,即高性能的候选区域孪生视觉跟踪器（SiamRPN）,验证了所提算法的有效性。     

An improvement in MSEPF for visual tracking

Recently, many approaches to applying a particle filter to a visual tracking problem have been proposed. However, it is hard to implement such a filter in a real-time system because it requires a great deal of computation time and considerable resources to achieve a high accuracy. In order to overcome this difficulty, especially the computation time, Shan and other workers have proposed combining a particle filter and mean shift in order to maintain the accuracy with a small number of particles. In their approach, the state of each particle moves to the point in the window with the highest likelihood value. It is known that the accuracy of an estimation depends on the size of the window, but a larger window size makes the computation slower. In this article, we propose a method for exploring the highest likelihood more quickly by means of random sampling. Moreover, the likelihood is also modified in terms not only of color cues, but also of motion cues for a greater accuracy in object tracking. The effectiveness of the proposed method is evaluated by real image sequence experiments.     

判别特征回归的鲁棒目标跟踪

目的 近年来,目标跟踪领域取得了很大进步,但是由于尺度变化,运动,形状畸变或者遮挡等造成的外观变化,仍然是目标跟踪中的一大挑战,因而有效的图像表达方法是提高目标跟踪鲁棒性的一个关键因素。方法 从中层视觉角度出发,首先对训练图像进行超像素分割,将得到特征向量集以及对应的置信值作为输入值,通过特征回归的方法建立目标跟踪中的判别外观模型,将跟踪图像的特征向量输入该模型,得到候选区域的置信值,从而高效地分离前景和背景,确定目标区域。结果 在公开数据集上进行跟踪实验。本文算法能较好地处理目标尺度变化、姿态变化、光照变化、形状畸变、遮挡等外观变化;和主流跟踪算法进行对比,本文算法在跟踪误差方面表现出色,在carScale、subway、tiger1视频中能取得最好结果,平均误差为12像素,3像素和21像素;和同类型的方法相比,本文算法在算法效率上表现出色,所有视频的跟踪效率均高于同类型算法,在carScale视频中的效率,是同类算法效率的32倍。结论 实验结果表明,本文目标跟踪算法具有高效性和鲁棒性,适用于目标发生外观变化时的目标跟踪问题。目前跟踪中只用了单一特征,未来考虑融合多特征来提升算法鲁棒性和准确度。     

基于多层特征增强的实时视觉跟踪

为了解决全卷积孪生视觉跟踪网络（SiamFC）出现相似语义信息干扰物使得跟踪目标发生漂移，导致跟踪失败的问题，设计出一种基于多层特征增强的实时视觉跟踪网络（MFESiam），分别去增强高层和浅层的特征表示能力，从而提升算法的鲁棒性。首先，对于浅层特征，利用一个轻量并且有效的特征融合策略，通过一种数据增强技术模拟一些在复杂场景中的变化，例如遮挡、相似物干扰、快速运动等来增强浅层特征的纹理特性；其次，对于高层特征，提出一个像素感知的全局上下文注意力机制模块（PCAM）来提高目标的长时定位能力；最后，在三个具有挑战性的跟踪基准库OTB2015、GOT-10K和2018年视觉目标跟踪库（VOT2018）上进行大量实验。实验结果表明，所提算法在OTB2015和GOT-10K上的成功率指标比基准SiamFC分别高出6.3个百分点和4.1个百分点，并且以每秒45帧的速度运行达到实时跟踪。在VOT2018实时挑战上，所提算法的平均期望重叠率指标超过2018年的冠军，即高性能的候选区域孪生视觉跟踪器（SiamRPN），验证了所提算法的有效性。