一种交通拥堵自动检测方法

目前,高速公路交通拥堵日趋频繁,然而大部分高速公路的管理方式仍然是通过人工查看轮巡监控视频来发现交通拥堵,效率低下。本文提出一种对高速公路轮巡监控视频自动检测交通拥堵的方法。该方法实时检测监控视频场景是否轮巡、切换、变动,自适应提取变动场景的道路边界结构并分类,然后在提取的道路边界结构范围内,计算宏观边缘占有率与加权平均光流速度两个宏观交通状态参数,并构造特征集,根据不同的道路边界类型选择对应支持向量机来实现对交通拥堵的自动检测和判别。实验结果表明,该方法能有效地对轮巡模式工作的高速公路监控视频检测交通拥堵,检测时间不大于30s,检测正确率达91.8%。     

基于图像熵的密集人群异常事件实时检测方法

在智能视频监控领域,为了提高密集人群中异常事件的检测效率,改善已有算法在实时性和适用性方面的不足,提出了一种实时高效的检测方法。该方法首先提取图像的全局光流强度作为运动特征,并构造全局光流强度的图像化表达;然后利用图像熵进行分析,获取正常状态下图像熵的统计参数;最后确定正常状态的可信区间和自适应的异常判定公式,从而判断异常事件是否发生。实验结果表明,该算法对尺寸为320×240像素的视频,平均每帧的检测时间低至0.031 s,且准确率可达96%以上,具有较高的检测效率,且实时性较好。     

基于光流法的卫星视频交通流参数提取研究

抽样分辨率达1米的高清卫星视频已经能够实现对地面较小的运动目标的实时监控。针对卫星视频中运动车辆目标仅显示为一个或几个像素点的特点,提出了一种基于光流法的卫星视频交通流参数提取的思路与方法。该方法利用卫星视频中车辆目标为像素点的特点,结合Shi-Tomasi角点检测方法实现车辆检测及车辆计数;在车辆检测的基础上利用光流法得到的连续视频帧中角点的位置信息进行双向车辆平均车速的计算,并对实验结果进行了对比分析。该文是基于卫星视频中小微运动车辆目标进行交通流参数提取的一次有益尝试。     

基于光流的视频缺陷检测及修复方法

为了修复视频中的划痕和斑点,提出一种基于光流的视频缺陷检测及修复方法。首先,根据光流场得到相邻帧对应像素之间的位置关系,利用对应点灰度差确定像素点所在位置是否为缺陷。其次,修正缺陷区域的光流,以修正光流指向的相邻帧修补点填补对应的缺陷点。最后,针对已修复的视频帧重新计算光流场并重复修复步骤,直到该帧满足迭代修复的收敛条件。针对DAVIS视频数据集的不同场景,模拟产生数量为单帧像素点总数1%左右的缺陷后进行检测修复实验,给出查全率与误识别率的关系曲线,其中,误识别率为0.1%时,查全率可达80%以上;修复后的SSIM大于0.991,LPIPS小于0.037。针对老旧视频的修复实验表明,算法能够有效去除细小划痕和大小斑块。     

适用于密集人群的异常事件实时检测方法

在密集人群场景下,针对现有异常检测算法在实时性和适用性方面的不足,提出了一种基于光流特征和卡尔曼滤波的实时检测方法。该方法首先提取图像的全局光流强度作为运动特征;然后对全局光流值进行卡尔曼滤波,并对残差进行分析;假设残差在正常状态下服从高斯分布,利用假设检验加以验证;运用最大似然(ML)估计得到残差的概率分布;在一定置信度下,确定正常状态的可信区间和异常状态的判定公式,并以此判断异常事件是否发生。实验结果表明,该方法对尺寸为320×240的视频,平均检测时间低至0.023 s/frame,且准确率可达95%以上。因而,该方法在保证较高检测率的同时,还具有良好的实时性。     

基于光流周期特性的视频帧率上转篡改检测

视频帧率上转是视频时域篡改的一种常见篡改手段,它通过周期性地在两个视频帧中间插入中间帧的方式,实现将视频由低帧率转换到高帧率的目标.本文提出了一种基于光流周期特性的视频帧率上转篡改检测算法,首先将视频转为帧图像序列,然后采用Horn-Schunck光流法计算每帧图像每个像素点的光流矢量,并计算相邻帧图像光流的变化率.最后利用快速傅里叶变换对光流变化率数据进行频谱分析,根据最高谱线的幅值与平均幅值的比值阈值来判别视频是否经过篡改.实验表明,算法不仅能够准确识别待测视频是否经过帧率上转篡改,并且提高了视频压缩的鲁棒性能,具有一定的实际应用价值.     

基于跟踪算法和模糊推理的交通事件检测方法

针对高速公路交通事故容易引发大规模拥堵的问题,提出了一种基于视频的快速检测算法,实现了事件信息的快速反馈。首先利用平稳序列法获取背景,并由背景差法获取前景。然后在使用凸包占有率对遮挡进行检测的基础上,采用改进的卡尔曼滤波特征匹配跟踪算法对车辆进行跟踪。最后通过对交通流的速度和流量进行检测,建立速度、流量与交通流状态之间的映射关系,运用模糊推理方法判别交通事件的发生。实验结果证明,本文提出的 方法能有效地获取前景信息,并能实时有效地对高速公路上的交通事件进行检测。     

基于光流模值变化频率的群体骚乱行为检测方法

群体骚乱行为突发性强、破坏性大,是视频监控关注的重点。研究群体骚乱行为的智能检测方法有助于提高视频监控的智能化水平。现有行为检测方法在检测群体骚乱行为时虚警率较高,实用性较差。因此,依据群体骚乱行为发生时光流模值变化大的特性,提出了一种基于光流模值变化频率的群体骚乱行为检测方法。该方法首先计算视频中每一帧图像上各像素点的光流;然后自适应求取反映像素点光流变化大小的二值映射图;接着分区块计算视频片段上的光流变化频率直方图,构建行为描述子;最后采用线性支持向量机进行特征训练与分类。实验表明,所提方法在检测群体骚乱行为时虚警率和漏警率低、识别率高,可广泛用于智能视频监控领域。     

多分辨率LK光流联合SURF的跟踪方法

针对交通监控中运动目标形变、雾霾天气、高速、光照不均、部分遮挡等复杂情况导致Lucas-Kanade（LK）算法跟踪不稳定问题,提出基于多分辨率LK光流算法联合快速鲁棒性特征（SURF）的跟踪算法。所提算法构建图像多分辨率小波金字塔,解决传统LK算法中同一像素点帧间大尺度运动易丢失问题;同时联合SURF尺度不变特征变换算法,提取特征点进行光流跟踪,并制定自适应模板实时更新策略;在减少光流计算量的同时增强运动目标抗复杂环境的能力。实验结果表明,新方法中特征点匹配准确快速,自适应性强,在交通复杂化境中跟踪稳定。     

基于Storm的高速公路实时交通指数评估方法的研究与实现

随着高速公路飞速的发展,出现的拥堵增加、监控困难等问题也日益严峻。首先本文在北京市城市交通拥堵指数的基础上,提出了利用交通指数对高速公路运行状况进行评估的方法,然后基于Storm大数据平台,实现了交通指数评估系统中的数据处理、分析、指数后台计算关键算法,最后通过手机APP、微信公众号等形式,实时地发布高速公路交通指数。实验结果表明：高速公路交通指数不仅可以反映整体的高速运行状况,也可以反映特定高速或者特定高速区间的实时拥堵情况,还可以实时监测高速路网的拥堵情况,为居民出行提供参考,对缓解道路拥堵起到积极作用。     

The Application of Video Semantics and Theme Representation in Automated Video Editing

This paper considers the automated generation of humorous video sequences from arbitrary video material. We present a simplified model of the editing process. We then outline our approach to narrativity and visual humour, discuss the problems of context and shot-order in video and consider influences on the editing process. We describe the role of themes and semantic fields in the generation of content oriented video scenes. We then present the architecture of AUTEUR, an experimental system that embodies mechanisms to interpret, manipulate and generate video. An example of a humorous video sequence generated by AUTEUR is described.     

VCMF: A Framework for Video Content Modeling

Current approaches to modeling the structure and semantics of video recordings restrict its reuse. This is because these approaches are either too rigidly structured or too generally structured and so do not represent the structural and semantic regularities of classes of video recordings. This paper proposes a framework which tackles the problem of reuse by supporting the definition of a wide range of models of video recordings and supporting reuse between them. Examples of the framework's use are presented and examined with respect to different kinds of reuse of video, current research, and the development of a toolset to support the framework.     

Model-Based Video Classification toward Hierarchical Representation,Indexing and Access

In this paper, we develop a content-based video classification approach to support semantic categorization, high-dimensional indexing and multi-level access. Our contributions are in four points: (a) We first present a hierarchical video database model that captures the structures and semantics of video contents in databases. One advantage of this hierarchical video database model is that it can provide a framework for automatic mapping from high-level concepts to low-level representative features. (b) We second propose a set of useful techniques for exploiting the basic units (e.g., shots or objects) to access the videos in database. (c) We third suggest a learning-based semantic classification technique to exploit the structures and semantics of video contents in database. (d) We further develop a cluster-based indexing structure to both speed-up query-by-example and organize databases for supporting more effective browsing. The applications of this proposed multi-level video database representation and indexing structures for MPEG-7 are also discussed.     

一种整体的视频匹配方法

给出一种视频时空配准的整体方法,提出一种视频内匹配与视频间匹配相结合的空间配准策略,改进动态时间扭曲方法以用于时间维的对齐.视频内匹配跟踪视频内各帧图像的特征点并记录其轨迹,视频间匹配配准不同视频的帧图像,使用轨迹对应提供图像配准所需的初始特征点对应,根据图像配准得到的特征点对应建立和更新轨迹对应.该匹配策略充分利用了视频的连贯性提高了匹配的稳定性和效率,同时提高了配准视频的连贯性.改进的动态时间扭曲方法通过极小化两段视频的整体距离建立视频之间的帧对应关系,保持视频内部各帧之间的时序关系并能处理非线性偏移     

Columbia's VoD and Multimedia Research Testbed with Heterogeneous Network Support

This paper reports our progress in developing an advanced video-on-demand (VoD) testbed, which will be used to accommodate various multimedia research and applications such as Electronic News on Demand, Columbia's Video Course Network, and Digital Libraries. The testbed supports delivery of MPEG-2 audio/video stored as transport streams over various types of networks, e.g., ATM, Ethernet, and wireless. Both software and hardware video encoders/decoders are used in the testbed. A real-time video pump and a distributed application control protocol (MPEG-2's DSM-CC) have been incorporated. Hardware decoders and set-tops are being used to test wide-area video interoperability. Our VoD testbed also provides an advanced platform for implementing proof-of-concept prototypes of related research. Our current research focus covers video transmission with heterogeneous quality-of-service (QoS) provision, variable bitrate (VBR) traffic modeling, VBR server scheduling, video over Internet, and video transmission over IP-ATM hybrid networks. An important aim is to enhance interoperability. Accommodation of practical multimedia applications and interoperability testing with external VoD systems has also been undertaken.     

Metadata-driven interactive web video assembly

The recent expansion of broadband Internet access led to an exponential increase of potential consumers of video on the Web. The huge success of video upload websites shows that the online world, with its virtually unlimited possibilities of active user participation, is an ideal complement to traditional consumption-only media like TV and DVD. It is evident that users are willing to interact with content-providing systems in order to get the content they desire. In parallel to these developments, innovative tools for producing interactive, non-linear audio-visual content are being created. They support the authoring process alongside management of media and metadata, enabling on-demand assembly of videos based on the consumer’s wishes. The quality of such a dynamic video remixing system mainly depends on the expressiveness of associated metadata. Eliminating the need for manual input as far as possible, we aim at designing a system which is able to automatically enrich its own media and metadata repositories continuously. Currently, video content remixing is available on the Web mostly in very basic forms. Most platforms offer upload and simple modification of content. Although several implementations exist, to the best of our knowledge no solution uses metadata to its full extent to dynamically render a video stream based on consumers’ wishes. With the research presented in this paper, we propose a novel concept to interactive video assembly on the Web. In this approach, consumers may describe the desired content using a set of domain-specific parameters. Based on the metadata the video clips are annotated with, the system chooses clips fitting the user criteria. They are aligned in an aesthetically pleasing manner while the user furthermore is able to interactively influence content selection during playback at any time. We use a practical example to clarify the concept and further outline what it takes to implement a suchlike system.

VisualGREP: A Systematic Method to Compare and Retrieve Video Sequences

Multimedia Tools and Applications - In this paper, we consider the problem of similarity between video sequences. Three basic questions are raised and (partially) answered. Firstly, at what...     

Detecting Roads in Stabilized Video with the Spatio-Temporal Structure Tensor

Video provides strong cues for automatic road extraction that are not available in static aerial images. In video from a static camera, or stabilized (or geo-referenced) aerial video data, motion patterns within a scene enable function attribution of scene regions. A “road”, for example, may be defined as a path of consistent motion — a definition which is valid in a large and diverse set of environments. The spatio-temporal structure tensor field is an ideal representation of the image derivative distribution at each pixel because it can be updated in real time as video is acquired. An eigen-decomposition of the structure tensor encodes both the local scene motion and the variability in the motion. Additionally, the structure tensor field can be factored into motion components, allowing explicit determination of traffic patterns in intersections. Example results of a real time system are shown for an urban scene with both well-traveled and infrequently traveled roads, indicating that both can be discovered simultaneously. The method is ideal in urban traffic scenes, which are the most difficult to analyze using static imagery.     

海量监控视频快速回放与检索技术

为解决海量监控视频的快速浏览和检索,介绍了一种基于目标索引的视频摘要和检索方法。该方法在光流分析的基础上,在画面的静止区域更新背景,运动的区域利用差分法分割出运动目标图像。经过优化的快速特征匹配和建立运动跟踪模型后,根据目标运动轨迹,按照时空距离进行聚类。在目标图像数据和运动参数进行XML结构化存储为索引的基础上,最后在检索时将符合条件的所有目标图像,按照其原有时间顺序逐帧贴到同一个背景图像中,形成动态的摘要视频。由于该方法剔除了背景中大量的时空冗余信息,可在较短回放时间内浏览全部有用目标,显著提高海量监控视频的查阅效率。     

基于VFW的远程视频捕获技术研究

多媒体技术在人们日常生活中的应用越来越广泛,图像、视频、音频等多媒体数据逐渐成为信息处理领域中主要的信息媒体形式。视频捕获技术是信息处理中的重要环节,研究该项技术具有重要的实用价值。文章提出一种基于VFW的远程视频捕获方法。该方法利用VFW捕获视频数据,采用H.263编码标准压缩视频数据,利用面向连接协议的流式套接字实现实时视频流的传输,结合多线程技术实现视频文件播放。然后,基于Windows操作系统设计实现了远程视频捕获系统。实验结果表明,该方法CPU占用率低、内存占用小,可靠性强,具有较好的应用价值。