面向多类别舰船多目标跟踪的改进CSTrack算法

针对海面舰船多目标跟踪过程中图像背景复杂、目标尺度差异大等难点,提出了一种改进CSTrack的舰船多目标跟踪算法。首先,针对CSTrack算法使用暴力解耦分解颈部特征造成目标特征损失的问题,提出了一种结合Res2net模块的改进互相关解耦网络RES_CCN,使网络解耦后获得更加细粒度的特征。其次,为提升对多类别舰船的跟踪性能,采用检测头网络解耦设计分别预测目标类别、置信度和位置。最后,采用MOT2016数据集进行消融实验,验证了所提模块的有效性,在新加坡海事数据集上进行测试,所提算法的多目标跟踪精度提升了8.4%,目标识别准确度提升了3.1%,优于ByteTrack等算法。本文所提算法具有跟踪精度高、误检率低等优点,适用于海面舰船多目标跟踪任务。

     

Tucker visual search-based hybrid tracking model and Fractional Kohonen Self-Organizing Map for anomaly localization and detection in surveillance videos

Anomaly detection (AD) in video is a challenging task employed in the intelligent video surveillance applications. This paper presents a technique for localizing and detecting anomalies in surveillance videos by proposing hybrid tracking model and Fractional Kohonen Self-Organizing Map (FKSOM). At first, the objects in the initial frames are detected by extracting the background and comparing with the succeeding frames. Then, a tracking model is developed to track the objects in the frame. Further, the features, such as object shape, speed, energy, correlation, and homogeneity, are extracted in the feature extraction process. Finally, the proposed FKSOM algorithm performs AD by identifying anomalous and normal events in the frame. The performance of the proposed technique is evaluated using the metrics, such as Multiple Object Tracking Precision (MOTP), accuracy, sensitivity, and specificity, where it obtains MOTP of 0.9895 with an average accuracy of 0.9339, the sensitivity of 0.9288 and specificity of 1.     

基于R-FCN框架的多候选关联在线多目标跟踪

在线多目标跟踪是实时视频序列分析的重要前提。针对在线多目标跟踪中目标检测可靠性低、跟踪丢失较多、轨迹不平滑等问题,提出了基于R-FCN网络框架的多候选关联的在线多目标跟踪模型。首先,通过基于R-FCN网络从KF预测结果和检测结果中获取更可靠的候选框,然后利用Siamese网络进行基于外观特征的相似性度量,实现候选与轨迹之间的数据关联,最后通过RANSAC算法优化跟踪轨迹。在人流密集和目标被部分遮挡的复杂场景中,提出的算法具有较高的目标识别和跟踪能力,大幅减少漏检和误检现象,跟踪轨迹更加连续平滑。实验结果表明,在同等条件下,与当前已有的方法对比,本文提出在目标跟踪准确度(MOTA)、丢失轨迹数(ML)和误报次数(FN)等多个性能指标均有较大提升。     

A multimodal approach for individual tracking of people and their belongings

Abstract

In this study, a fully automatic surveillance system for indoor environments which is capable of tracking multiple objects using both visible and thermal band images is proposed. These two modalities are fused to track people and the objects they carry separately using their heat signatures and the owners of the belongings are determined. Fusion of complementary information from different modalities (for example, thermal images are not affected by shadows and there is no thermal reflection or halo effect in visible images) is shown to result in better object detection performance. We use adaptive background modeling and local intensity operation for object detection and the mean-shift tracking algorithm for fully automatic tracking. Trackers are refreshed to resolve potential problems which may occur due to the changes in object’s size, shape and to handle occlusion-split and to detect newly emerging objects as well as objects that leave the scene. The proposed scheme is applied to the abandoned object detection problem and the results are compared with the state of art methods. The results show that the proposed method facilitate individual tracking of objects for various applications, and provide lower false alarm rates compared to the state of art methods when applied to the abandoned object detection problem.     

Fast image foreground object segmentation based on block texture for embedded system implementation

In this paper, a novel block-based foreground object detection method based on block texture is presented. It can significantly reduce the memory usage when constructing the background model in dynamic scenes. The proposed background model and detection algorithm are suitable for implementing on embedded system platforms with resource limitations. The experimental results of processing benchmark videos show that our method has outcomes that are very close to ground truth segmentation. In addition, the proposed method requires approximately 23.97% less memory than the latest algorithms. Finally, the proposed approach is implemented on an embedded system platform. The processing speed can achieve a real-time rate of at least 20 fps, which is an improvement of 17.64% as compared to the latest algorithms.     

Automatic skin segmentation and tracking in sign language recognition

Skin segmentation and tracking play an important role in sign language recognition. A framework for segmenting and tracking skin objects from signing videos is described. It mainly consists of two parts: a skin colour model and a skin object tracking system. The skin colour model is first built based on the combination of support vector machine active learning and region segmentation. Then, the obtained skin colour model is integrated with the motion and position information to perform segmentation and tracking. The tracking system is able to predict occlusions among any of the skin objects using a Kalman filter (KF). Moreover, the skin colour model can be updated with the help of tracking to handle illumination variation. Experimental evaluations using real-world gesture videos and comparison with other existing algorithms demonstrate the effectiveness of the proposed work.     

Moving object detection and tracking in videos through turbulent medium

This paper addresses the problem of identifying and tracking moving objects in a video sequence having a time-varying background. This is a fundamental task in many computer vision applications, though a very challenging one because of turbulence that causes blurring and spatiotemporal movements of the background images. Our proposed approach involves two major steps. First, a moving object detection algorithm that deals with the detection of real motions by separating the turbulence-induced motions using a two-level thresholding technique is used. In the second step, a feature-based generalized regression neural network is applied to track the detected objects throughout the frames in the video sequence. The proposed approach uses the centroid and area features of the moving objects and creates the reference regions instantly by selecting the objects within a circle. Simulation experiments are carried out on several turbulence-degraded video sequences and comparisons with an earlier method confirms that the proposed approach provides a more effective tracking of the targets.     

基于形状信息的Bayes分类方法

本文提出了一种新的基于形状信息的Bayes分类方法,以实现对图像中单个物体的分类。该方法首先运用图像边缘提取和配准算法,构造一个形状相似性能量泛函,并利用其计算形状信息的先验概率。然后,结合图像中物体其它特征的后验概率,通过Bayes方法进行分类。本文将该方法应用于一个病原菌图像分类的实际问题,实验结果表明,该方法是十分有效的,不仅降低了分类所需的特征维数,而且提高了分类精度,并能满足实际问题中所要求的计算速度。     

一种新的Mean-Shift对象跟踪方法

由于存在外点干扰,传统的Mean-Shift对象跟踪方法不能有效的跟踪尺寸逐步变大的目标.猜想利用对象初始模型能有效排除外点对跟踪结果的干扰,在此基础上本文提出了一种利用对象初始模型的候选目标模型和相似度测量方法;为了使内核窗口中心及大小与对象形心及大小一致,根据内核窗口外一定宽度子带内像素分布情况,动态调整内核窗口大小和位置.通过对尺寸逐渐变大的汽车进行跟踪,验证了算法的有效性,同时试验结果也证明新算法具有更低的计算复杂度.     

结合颜色和结构信息的粒子滤波跟踪算法

单纯依靠颜色信息往往使得目标跟踪不稳定.为了在复杂背景和光照变化等干扰情况下,能够准确地跟踪到感兴趣的目标,本文提出了将颜色和结构信息相结合的方法.该方法在基于颜色直方图的粒子滤波跟踪框架中,利用目标的灰度图像建立结构模型,通过结构相似性质量因子衡量目标在两帧图像之间的相似性.实验表明,该算法在使用相同粒子数目的情况下较传统的基于颜色直方图的粒子滤波跟踪算法鲁棒性更高,可以用于特定场合可疑目标的跟踪.     

一种视频图像序列中运动对象的分割与跟踪算法

本文提出了一种视频图像序列中运动对象的分割与跟踪算法。该算法通过Canny算子检测出差帧图像的边缘信息,并结合当前帧与背景帧的边缘图像,提取出运动对象。在后续帧中通过建立前帧感兴趣运动对象与当前帧中各运动对象的帧间向量来跟踪当前帧中感兴趣的视频对象。实验结果表明,该算法可行,而且由于该算法简单、计算复杂度小,能很好地满足实时监控系统中对感兴趣运动对象的提取与跟踪。     

基于演化非对称核函数的均值漂移跟踪算法

针对传统均值漂移跟踪算法中采用的对称核函数模板中包含了较多背景像素点,影响跟踪精确度和稳定性的缺点,在固定非对称核函数的基础上对均值漂移跟踪算法进行了改进,提出了一种基于演化非对称核函数的均值漂移目标跟踪算法。本文算法首先介绍了将非对称核函数模板引用到均值漂移算法框架的关键问题——模板中心——的计算方法;其次将非对称核函数模板的表述和演化有机结合,提出了利用区域相似度的目标轮廓水平集演化算法并阐述了非对称核函数模板的更新策略。实验结果表明,相比现有的方法,本文提出的基于演化非对称核函数模板均值漂移跟踪算法具有更高的准确性和可靠性,同时也能满足一般目标跟踪任务的实时性要求。     

基于双层粒子滤波和半监督Hough Forests 的多目标跟踪

本文针对单目摄像头、复杂可变背景环境下的多目标跟踪问题,将 tracking-by- detection 方法与粒子滤波相结合,从不稳定的信息源中提取高置信度模型作为观测,在半监督学习框架中实现了动态视频场景中的多个目标跟踪,并设计了一个多目标的维护机制以应对遮挡、背景变化、目标进出场景等可能引起目标混淆的情况.实验证明,本文提出的算法能够稳定跟踪复杂场景中的多个目标,有效区分不同目标,对目标的遮挡、背景干扰等均有良好的处理效果.     

结合水平集和粒子滤波的人脸轮廓跟踪

针对视频序列中的人脸轮廓跟踪问题,提出了一种将改进水平集方法引入到粒子滤波框架中的轮廓跟踪算法.此算法使用零水平集函数对人脸轮廓进行表示,通过水平集函数的演化对人脸轮廓进行逼近.为了解决人脸遮挡问题,将形状先验加入到演化过程中来约束曲线的演化.算法在粒子滤波的框架下可以同时跟踪人脸区域的仿射运动和人脸轮廓的形变.实验结果表明,所提出的'方法可以对人脸轮廓进行精确的跟踪,并对外界的光照变化,背景干扰,人脸部分遮挡有较强的鲁棒性.     

基于假设理论的序列图像中多目标检测和跟踪

本文提出一种基于目标检测的多维假设多目标检测和跟踪方法,此算法对于序列图像的照明变化和遮蔽现象具有很高的鲁棒性.首先,对序列图像进行背景抑制、时域滤波和杂波剔除的预处理,得到单帧的初始目标检测结果.对于初始检测中存在的漏检和误检现象,采用基于假设理论的跟踪概率模型优化初始检测轨迹;将目标的跟踪信息反馈于目标检测模块,形成一闭环自适应跟踪系统,达到多目标的最优检测和跟踪.实验结果表明了所提出的方法在多目标检测和跟踪中的可行性和有效性.     

基于ReInspect算法的多目标追踪

为了提高复杂场景下多目标检测的准确性,提出了一种基于ReInspect算法的对于多个运行目标的检测方法.该算法基于OverFeat算法和Faster R-CNN算法的思想,加入LSTM(long short-term memory)循环网络结构用于记录多个目标的特征序列;通过调整LSTM网络特征标签信息,预处理损失函数...     

Detection of circlelike overlapping objects in thermal spray images

In this paper, we present a new algorithm for the detection of distorted and overlapping circlelike objects in noisy grayscale images. Its main step is an edge detection using rotated difference kernel estimators. To the resulting estimated edge points, circles are fitted in an iterative manner using a circular clustering algorithm. A new measure of similarity can assess the performance of algorithms for the detection of circlelike objects, even if the number of detected circles does not coincide with the number of true circles. We apply the algorithm to scanning electron microscope images of a high-velocity oxygen fuel (HVOF) spray process, which is a popular coating technique. There, a metal powder is fed into a jet, gets accelerated and heated up by means of a mixture of oxygen and fuel, and finally deposits as coating upon a substrate. If the process is stopped before a continuous layer is formed, the molten metal powder solidifies in form of small, almost circular so-called splats, which vary with regard to their shape, size, and structure and can overlap each other. As these properties are challenging for existing image processing algorithms, engineers analyze splat images manually up to now. We further compare our new algorithm with a baseline approach that uses the Laplacian of Gaussian blob detection. It turns out that our algorithm performs better on a set of test images of round, spattered, and overlapping circles.     

An Efficient Crossing-Line Crowd Counting Algorithm with Two-Stage Detection

Crowd counting is a challenging task in crowded scenes due to heavy occlusions, appearance variations and perspective distortions. Current crowd counting methods typically operate on an image patch level with overlaps, then sum over the patches to get the final count. In this paper we describe a real-time pedestrian counting framework based on a two-stage human detection algorithm. Existing works with overhead cameras is mainly based on visual tracking, and their robustness is rather limited. On the other hand, some works, which focus on improving the performances, are too complicated to be realistic. By adopting a line sampling process, a temporal slice image can be obtained for pedestrian counting without the need for visual tracking. Only ten low level features are extracted from the input image to establish a feature vector. As a result, our algorithm is more efficient and accurate than existing methods. Pedestrians in the temporal slice image are then located by the two-stage detection algorithm, which is largely based on support vector machine and affinity propagation clustering. Moreover, a novel algorithm is proposed to determine the moving directions of pedestrians by comparing the centers of them in two temporal slice images. Extensive experiments reveal that our system achieves satisfaction performances in terms of both robustness and efficiency.     

快速运动目标的Mean shift跟踪算法

针对Mean shift本身的理论缺陷,提出Mean shift和卡尔曼滤波器相结合的快速目标跟踪算法。利用卡尔曼滤波器来获得每帧Mean shift算法的起始位置,然后再利用Mean shift算法得到跟踪位置。在目标出现大比例阻挡情况时,利用卡尔曼残差的计算来关闭和打开卡尔曼滤波器,此时,目标位置的线性预测替代了卡尔曼的作用。试验证明,本算法可以实现对快速运动目标的跟踪,对阻挡也有很好的鲁棒性。     

Frontal gait recognition based on spatio-temporal interest points

This paper proposes a spatio-temporal interest point (STIP)-based human gait recognition method for frontal gait videos. The proposed method extracts histograms of oriented gradient (HOG) for STIPs directly from gait videos without involving a gait cycle segment, which is required by silhouette analysis as used by most gait recognition methods. Moreover, silhouette extraction may be affected by noise and carried objects leading to consequent recognition error. Matches of STIPs between two gait videos are then found in terms of HOG to measure the similarity of two videos so as to achieve the goal of gait recognition. The experimental results offer evidence that our method outperforms the existing methods on the Carnegie-Mellon University Motion of Body database and the Chinese Academy of Sciences Institute of Automation data-set B.