Integrated video object tracking with applications in trajectory-based event detection

This work presents an automated and integrated framework that robustly tracks multiple targets for video-based event detection applications. Integrating the advantages of adaptive particle sampling and mathematical tractability of Kalman filtering, the proposed tracking system achieves both high tracking accuracy and computational simplicity. Occlusion and segmentation error cases are analyzed and resolved by constructing measurement candidates via adaptive particle sampling and an enhanced version of probabilistic data association. Also, we integrate the initial occlusion handling module in the tracking system to backtrack and correct the object trajectories. The reliable tracking results can serve as the foundation for automatic event detection. We also demonstrate event detection by classifying the trajectories of the tracked objects from both traffic monitoring and human surveillance applications. The experimental results have shown that the proposed tracking mechanism can solve the occlusion and segmentation error problems effectively and the events can be detected with high accuracy.     

Performance measures for video object segmentation and tracking

We propose measures to evaluate quantitatively the performance of video object segmentation and tracking methods without ground-truth (GT) segmentation maps. The proposed measures are based on spatial differences of color and motion along the boundary of the estimated video object plane and temporal differences between the color histogram of the current object plane and its predecessors. They can be used to localize (spatially and/or temporally) regions where segmentation results are good or bad; and/or they can be combined to yield a single numerical measure to indicate the goodness of the boundary segmentation and tracking results over a sequence. The validity of the proposed performance measures without GT have been demonstrated by canonical correlation analysis with another set of measures with GT on a set of sequences (where GT information is available). Experimental results are presented to evaluate the segmentation maps obtained from various sequences using different segmentation approaches.     

基于分水岭分割和尺度不变特征点的多目标全自主跟踪算法

该文针对多目标的鲁棒跟踪问题,设计了一种基于图像分水岭分割和尺度不变特征变换(SIFT)特征点的多目标全自主跟踪算法。为规避图像平坦区域,提出在原图上叠加规则坡度图的思想,并在浮点域进行一定尺度高斯模糊处理,将区域极小值点作为种子点完成分水岭分割,并将极值点作为目标特征点,通过前后帧分水岭映射生成特征点短时轨迹,自动检测运动目标。之后依据目标所处状态(是否发生遮挡)和分水岭分割图建立、更新目标SIFT特征池,结合分水岭映射、SIFT特征池匹配完成对目标的鲁棒跟踪。实验结果表明,该算法能有效完成视频中多目标的持续跟踪,并对目标遮挡有较好的鲁棒性。     

局部感知下的稀疏优化目标跟踪方法

针对传统稀疏表示跟踪算法在复杂背景中易出现跟踪漂移问题,该文提出一种局部感知下的稀疏优化目标跟踪方法。首先,将首帧确定的目标区域进行非重叠均匀分割,并利用目标的全局特征和局部特征联合建模。然后,提出一种局部感知校验方法约束稀疏优化匹配过程,从而确定最优匹配样本。最后,在模板更新中提出一种决策方法对遮挡进行检测,并针对不同遮挡情况采取相应的更新策略,使得更新后的模板集更加完善。实验在10个标准库视频序列中测试,并与目前较流行的目标跟踪算法在跟踪效果、成功率等方面进行比较,实验结果表明,提出的跟踪方法在局部遮挡、目标形变、复杂背景等条件下跟踪准确、适应性强。     

基于双步相关滤波的目标跟踪算法

针对目标跟踪中因目标遮挡、目标出视野导致的跟踪失败问题,为提高相关滤波目标跟踪算法的鲁棒性,提出了一种基于双步相关滤波的目标跟踪算法。首先根据方向梯度直方图特征不同单元大小目标表征的特点,提出双步相关滤波目标跟踪框架,在提高目标跟踪精度的同时保证了跟踪速度;然后融合多种目标特征,来获得目标更加全面的特征表征,以提高目标跟踪的鲁棒性;最后提出基于目标跟踪置信度指标的目标模板自适应更新策略,来解决目标遮挡时目标模板被污染的问题。实验在OTB100标准目标跟踪数据集上进行验证,通过与其他跟踪算法进行比较结果表明,该算法与其中最优跟踪算法相比,目标跟踪精度提升6.0%,目标跟踪成功率提升5.5%,平均跟踪速度为27.4 fps,保证了目标跟踪的实时性。实际目标跟踪应用中,在目标严重遮挡等情况下,该算法仍然可以对目标进行稳定精确地跟踪。     

基于模板匹配的视频对象分割

视频对象分割是MPEG-4标准关键技术.本文结合模板匹配和基于运动估值和补偿的对象跟踪方法,提出了一种可以从复杂场景中分割出MPEG-4视频对象的新方法.在使用运动估值和补偿得到分割掩膜后,以初始帧对象颜色为模板,在当前帧的轮廓边界区域通过模板匹配检测对象,使轮廓精确化.本文方法在一定范围内有效解决了遮挡问题,并能够以初始帧跟踪任意长序列中的对象.     

颜色不变量空间下的行人跟踪算法研究

为了有效的克服遮挡问题准确跟踪行人,本文提出了一种通过不断学习新的外观模型来自适应跟踪行人的跟踪算法。该算法首先将颜色不变量特征平面作为根特征来表示初始特征空间;然后将跟踪问题转化为0或1的二进制问题,通过局部最小二乘法（PLS）来对目标外观特征和对应的类型标签进行建模得到前景和背景的模板。随着目标外观的变化,本文利用局部最小二乘法（PLS）在颜色不变量平面上分析多个外观特征的样本信息,不断的更新模板,从而达到对遮挡具有很好鲁棒性的行人跟踪效果。通过对通用数据集进行试验表明:该算法在颜色暗淡和颜色鲜明的视频图片中都能达到很好的跟踪效果。      

基于改进MeanShift的目标跟踪算法

针对传统Meanshift算法在某些干扰或遮挡情况下不能保证跟踪的准确性,以及目标模型内的背景像素也会造成定位偏差的问题,提出一种基于MeanShift的改进算法。首先对目标模型进行改进,通过目标与背景的区分度引入权系数,在目标模型中进行加权处理,可达到降低目标模型内背景像素对跟踪定位精度的影响。然后,将跟踪窗进行分块,对各子块使用改进目标模型的Meanshift算法进行跟踪。最后,用匹配度最大的两个子块加权决定目标的最终位置,从而在目标发生遮挡时能有效剔除被遮挡子块对目标定位的影响。实验表明,在复杂背景下,新算法仍然可以有效、准确地跟踪运动目标。     

一种鲁棒的基于在线boosting目标跟踪算法研究

在线boosting的优点在于将跟踪问题看做分类问题,并且根据目标的变化实时选择相应的目标特征来进行跟踪。但该方法存在的主要问题是由于遮挡等情形的存在,目标特征的每次变换可能引入少量错误,长时间执行后,错误的积累将导致跟踪位置的漂移。实验发现,在线boos-ting方法中选择器权重在全局中的传递导致了漂移的发生,而现实跟踪问题中遮挡时只影响局部区域,而不是全局区域。对于这个问题,本文提出了结合在线boosting和分块的方法来解决这个问题。此boosting算法中选择器权重的变化仅在本块中,而不是全局传递,从而避免了错误在全局的累积,进而避免了漂移问题的产生。本文的方法通过跟踪各种目标视频序列实验表明,即使对于严重遮挡也具有很强的鲁棒性,同时也能实时的跟踪目标,即每秒跟踪目标量超过10个。     

Kernelized correlation tracking based on point trajectories

Visual tracking is one of the most important directions in computer vision.However,many state-of-the-art algorithms cannot track the interested object reliably due to occlusion during tracking process,which leads to deficiency of object information.In order to solve occlusion problem,a kernelized correlation tracking method based on point trajectories was proposed.Through analyzing long-term motion cues of the local information,point trajectories were labeled by spectral clustering.These labeled points were used to differentiate the foreground and background objects and thus detect whether the target was occluded or drifts.If drifting and occlusion occur,re-detection was used to detect the re-entering of the target.Experimental results show that the proposed algorithm can handle occlusion and drifting problems effectively.     

Entropy-based metrics for the analysis of partial and total occlusion in video object tracking

Metrics measuring tracking reliability under occlusion that are based on mutual information and do not resort to ground truth data are proposed. Metrics for both the initialisation of the region to be tracked as well as for measuring the performance of the tracking algorithm are presented. The metrics variations may be interpreted as a quantitative estimate of changes in the tracking region due to occlusion, sudden movement or deformation of the tracked object. Performance metrics based on the Kullback-Leibler distance and normalised correlation were also added for comparison purposes. The proposed approach was tested on an object tracking scheme using multiple feature point correspondences. Experimental results have shown that mutual information can effectively characterise object appearance and reappearance in many computer vision applications.     

一种基于改进Chan-Vese模型的图像分割方法

为了解决红外目标在运动过程中因遮挡带来的误分割和误跟踪问题,在参考Chunming Li提出的无需重新初始化水平集方法的基础上,提出了基于改进Chan-Vese模型的图像分割方法,给出了模型的能量函数及数值实现,通过对实际采集红外序列图像数据的对比实验验证了所提图像分割模型在解决目标受到遮挡问题中的有效性.     

基于特征子空间的自适应多视角目标跟踪算法

提出了一种新的自适应特征子空间跟踪算法,该算法通过计算跟踪目标的似然来自适应调整模型更新的权重,以减小更新过程中样本误差积累导致的模型漂移.同时,跟踪算法利用多视角贝叶斯理论框架进行多视角的信息融合,并对跟踪模型进行分块处理和更新,以提高跟踪精确度.仿真结果表明,本算法比对比算法的跟踪误差更小,并能够更好地解决目标遮挡和形变等问题,从而得到精确、高效的跟踪结果.     

一种改进的核相关滤波目标跟踪方法研究

针对目标跟踪领域视频图像序列经常出现遮挡、光照变化和快速移动等难点,提出了一种融合目标上下文信息和超像素级特征的核相关滤波跟踪方法,目标与其周围的信息存在相关性,利用中层视觉特征(超像素)构建目标的上下文滤波模型,在相关滤波框架下,实现对目标的快速训练和定位。实验结果表明,和其他相关滤波类算法相比,所提出的算法在处理以上跟踪难点时,精确度更高,且鲁棒性更强,同时能以较快的速度运行,满足实时性的要求。     

基于相关滤波器的视觉目标跟踪方法

为了解决视觉目标跟踪中的尺度预测的难题,本 文在核相关滤波器的目标跟踪的框架下给出了一种 尺度估计策略,并对传统的核相关跟踪方法中目标模型的在线更新方法进行了修改,提出了一 种多尺度视觉 目标跟踪算法。首先,通过对正则化最小二乘分类器(RLS)学习获得位置和尺度核相关滤波 器(KCF);然后,寻找位 置和尺度KCF输出响应的最大值,完成目标位置和尺度的检测;最后,在线更新目 标模型。实验中,对 12组具有挑战性的标准视频序列进行测试。实验结果表明,相对于现 有的3种基于相关滤波器的跟 踪方法中的最优者,本文方法的平均中心位置误差(CLE)减 少了7.0pixels,平均成功率(SR)提 高了18.3%,平均距离(DP)精度提高了 5.6%;在目标发生尺度、光照、姿态变化、部 分遮挡、旋转及快速运 动等复杂情况下,本文方法均有较强的适应性,具有重要的理论和应用研究价值。     

基于可变形模型的目标跟踪算法

近年来目标跟踪技术的研究已经有了很大的进展,但目标的遮挡和形变仍然是目标跟踪算法面临的重大挑战。针对这些问题提出了一种基于可变形模型的目标跟踪算法。首先,利用可变形模型对跟踪目标进行表达,该模型将目标分为若干子块,目标的特征由局部子块特征和全局特征共同构成。将目标的特征和子块之间的空间关系结合起来,给出了对目标的一个统一的相似度度量函数。然后,在线训练一个结构化输出支持向量机作为分类器,该分类器的输出是可变形模型中目标的结构化描述。利用该分类器可以在视频及图像序列中准确地检测到目标,完成跟踪。通过实验比较,该算法的跟踪性能优于其他主流跟踪算法,尤其在目标发生遮挡和形变的时候仍能准确跟踪。     

一种红外和可见光双通道视频目标跟踪方法

室外场景下由于场景背景条件变化容易导致视频目标跟踪稳定性差。该文提出一种利用红外和可见光传感器的双通道视频目标跟踪方法。该算法利用可见光图像的目标颜色特征和红外图像的目标轮廓特征,结合均值漂移算法与水平集曲线演化实现目标定位,并给出了目标尺度和模板更新方法;对多目标跟踪的互相遮挡问题,通过判断目标合并与分离实现遮挡时多个目标的定位。实验结果表明,该文方法能够有效处理光照变化、阴影、遮挡等情况,实现目标的稳定跟踪。     

Structural sparse coding seeds–active appearance model for object tracking

In this paper, we propose a tracking algorithm that can robustly handle appearance variations in tracking process. Our method is based on seeds–active appearance model, which is composed by structural sparse coding. In order to compensate for illumination changes, heavy occlusion and appearance self-updating problem, we proposed a mixture online learning scheme for modeling the target object appearance model. The proposed object tracking scheme involves three stages: training, detection and tracking. In the training stage, an incremental SVM model that directly measures the candidates samples and target difference. The proposed mixture generate–discriminative method can well separate two highly correlated positive candidates images. In the detection stage, the trained weighted vector is used to separate the target object in positive candidates images with respect to the seeds images. In the tracking stage, we employ the particle filter to track the object through an appearance adaptive updating algorithm with seeds–active constrained sparse representation. Based on a set of comprehensive experiments, our algorithm has demonstrated better performance than alternatives reported in the current literature.     

多模型融合的孪生网络视觉跟踪

为提升孪生网络视觉跟踪算法的准确性,提出一种融合多任务差异化同质型模型的孪生网络视觉跟踪算法。首先在决策层对孪生网络视觉跟踪模型与目标分割模型进行融合,然后结合多尺度搜索区域、目标上下文特征、多学习率模型更新策略进行跟踪。在标准数据集VOT、OTB、LaSOT、UAV123上进行算法评估。实验结果表明,所提算法在遮挡、快速运动、光照变化等干扰下可以稳定跟踪目标。