基于视频的静止虚拟摄像机生成

虚拟摄像机的生成是虚拟广告系统中一项关键技术,提出一种静止虚拟摄像机生成方法。首先,在视频图像中自动检测比赛场地的特征点,然后根据特征点在图像坐标系与世界坐标系上的匹配关系,利用摄像机定标技术计算出真实摄像机的内外参数,接着把摄像机的内参数和外参数分别转换为虚拟摄像机的视点转换矩阵和投影矩阵,最后在此虚拟摄像机下对虚拟场景进行渲染,把渲染出的图像插入到视频中。实验结果表明该方法简单、有效、易实现。     

Registering real and virtual imagery

Registering a virtual scene with a real scene captured by a video camera has a number of applications including visually guided robotic navigation, surveillance, military training and operation. The fundamental problem involves several challenging research issues including finding corresponding points between the virtual and the real scene and camera calibration. This paper presents our research in defining and mapping a set of reliable image features for registering the two imageries, extracting and selecting reliable control points for the construction of intrinsic and extrinsic camera parameters. A number of innovative algorithms are presented followed by extensive experimental analysis. An application of registering virtual database image with video image is presented. The algorithms we developed for calculating and matching linear structured features and selecting of reliable control points are applicable to image registration beyond virtual and real imageries.     

深度学习及其在图像物体分类与检测中的应用综述

传统的图像物体分类与检测算法及策略难以满足图像视频大数据在处理效率、性能和智能化等方面所提出的要求。深度学习通过模拟类似人脑的层次结构建立从低级信号到高层语义的映射,以实现数据的分级特征表达,具有强大的视觉信息处理能力,成为应对这一挑战的前沿技术和国内外研究热点。首先论述了深度学习的起源、发展历程及理论体系;然后分别围绕图像物体分类和检测,总结了近年来深度学习在视觉领域的发展;最后对深度学习及其在视觉领域目前存在的诸多问题以及后续的研究方向进行了分类探讨。     

基于圆形靶标的数码相机外部参数标定算法

数码相机的外部参数标定是应用计算机视觉进行现场测量和图像处理的关键技术。提出一种基于圆形靶标的数码相机外部参数标定算法,首先提出一种用于估计靶标上圆的圆心在该相机像平面的像坐标的迭代学习算法,在此基础上,利用已知的数码相机内部参数和成像几何模型计算出数码相机的外部参数。实验结果表明算法具有较高的精度和良好的稳定性。     

Panoramic video stitching from commodity HDTV cameras

Digital camera and smartphone technologies have made high-quality images and video pervasive and abundant. Combining or stitching collections of images from a variety of viewpoints into an extended panoramic image is a common and popular function for such devices. Extending this functionality to video however, poses many new challenges due to the demand for both spatial and temporal continuity. Multi-view video stitching (also called panoramic video stitching) is an emerging, common research area in computer vision, image/video processing and computer graphics and has wide applications in virtual reality, virtual tourism, surveillance, and human computer interaction. In this paper, we study and solve the major technical and practical problems in the complete process of stitching a high-resolution multi-view video into a high-resolution panoramic video. The challenges addressed include video stabilization, efficient high-definition multi-view video alignment and stitching, color correction, and blurred frame detection and repair. The proposed approaches have been successfully applied in a high-quality virtual reality system—the Virtual Exercise Environment (VEE) system.     

基于三维全景图的校园漫游系统王绪宛,雷蕾

虚拟现实是利用电脑模拟产生一个三维空间的虚拟世界,让使用者如同身历其境,而三维全景图是虚拟现实中一种重要的场景表示方法,生成三维全景图最简单的方法是直接使用全景相机等工具拍摄全景图,由于相机设备价格昂贵,限制了普遍使用;基于图像拼接技术对图像序列进行处理后生成全景图的低成本方法。该方法在南阳理工学院虚拟校园系统中使用,实现了虚拟校园的实景漫游,大大降低了成本。     

图像超分辨率深度学习研究及应用进展

图像超分辨率重建是用低分辨率图像重建出对应的高分辨率图像的过程。目前,图像超分辨率技术已经成功应用于计算机视觉和图像处理领域。近年来,由于深度学习具有能够从大量数据中自动学习特征的能力,因此被广泛应用于图像超分辨率领域中。介绍了图像超分辨重建的背景,详细总结了用于图像超分辨率的深度学习模型,阐述了图像超分辨率技术在卫星遥感图像、医学影像、视频监控、工业检测任务方面的应用。总结了图像超分辨算法的当前研究现状以及未来发展方向。     

旋转式电磁机构动态位移的图像测试技术

将一种基于高速摄像机和虚拟仪器的动态测试图像处理技术运用于旋转式电磁机构的动态位移测试。在LabVIEW虚拟仪器的平台上,运用图像处理技术对旋转式电磁电器动态过程连续的高速图像进行处理与分析,通过提取触头运动过程中标记点的坐标,绘制了电磁电器动态机构的位移曲线。     

A hemispherical imaging camera

Most of the existing panoramic cameras have cylindrical field of view (FOV) which has a 360° width in azimuth, but a limited height in elevation. However, a hemispherical (or near spherical) FOV is highly desirable in many applications such as robot navigation and surveillance. This paper presents a planar-mirror-based panoramic camera (HEM) which is capable of acquiring nearly hemispherical FOV panoramic images at high and substantially uniform resolution, from a single viewpoint, at video rates. The camera consists of multiple imaging sensors and a hexagonal prism made of planar mirror faces. The sensors are positioned in such a way that they image different parts of the scene from the same virtual viewpoint, either directly or after reflections off the prism. A panoramic image is constructed by concatenating the images taken by different sensors. The proposed system is designed such that the resolution across entire FOV has the same level of uniformity as delivered by a conventional, non-panoramic camera over a small FOV, and the sensor area utilization is maximized. We also present a novel camera placement technique that helps co-locate the optic points of all seven sensors at a single viewpoint, and a method to compensate for artifacts near mirror boundaries caused by finite size of lens aperture. An implementation of the proposed system as well as resulting hemispherical panoramic images are included.     

一种适用于动态场景的运动目标提取新算法

在动态场景中提取运动目标是开展视频分析的关键问题,也是当前计算机视觉与图像处理技术领域中的热门课题。本文提出了一种适用于动态场景的运动目标提取新算法,算法先根据摄像机全局运动模型计算全局运动参数,再利用三帧差分法得到分割的前景。将分割为背景的像素点映射到邻近帧,求得各帧的像素点为背景时其高斯模型的均值及方差。最后利用粒子滤波预测出下一帧前景区域,计算各像素点为前景的概率,获得运动目标的视频分割结果。实验表明,本文算法有效地克服了由于全局运动模型参数估算偏差而导致的累积误差,能以更高精度实现跳水运动视频中的目标分割。     

纹理映射中的平面校正技术研究

为了快速实时地进行由平面组成的结构景物的3D建模问题,文中介绍了一种在进行图像3D重构时纹理映射中的平面校正方法。介绍了三角形模型在图像处理、图形绘制、虚拟现实等技术中的重要作用。从射影几何的角度出发,给出了从两幅视图进行景物三维重构的分层重构方法。在已知欧氏重构即摄像机内参数的基础上,介绍一种基于标定的平面射影失真矫正方法。通过此方法,将矫正过的纹理映射到欧式点重构结构中,得到景物的3D模型。经实验验证,这种方法在处理由平面组成的景物的3D重构中是实时有效的。     

The Big Data Analysis on the Camera-based Face Image in Surveillance Cameras*

In the Big-Data era, currently how to automatically realize acquisition, refining and fast retrieval of the target information in a surveillance video has become an urgent demand in the public security video surveillance field. This paper proposes a new gun-dome camera cooperative system, which solves the above problem partly. The system adopts a master-slave static panorama-variable view dualcamera cooperative video-monitoring system. In this dual-camera system the gun camera static camera） with a wide viewing -angle lenses is in charge of the pedestrian detection and the dome camera can maneuver its focus and cradle orientation to get the clear and enlarged face images. In the proposed architecture, Deformable Part Model (DPM) method realizes real-time detection of pedestrians. The look-up table method is proved feasible in a dual-camera cooperative calibration procedure, while the depth information of the moving target changes slightly. As respect to the face detection, the deep learning architecture is exploited and proves its effectiveness. Moreover, we utilize the Haar-Like feature and LQV classifier to execute the frontal face image capture. The experimental results show the effectiveness and efficiency of the dual-camera system in close-up face image acquisition.     

一种新型的嵌入式智能图像传感器设计

提出并设计一种新型的嵌入式智能图像传感器,该传感器具备一定的室外场景(主要为道路和公共场所)状态感知能力,适用于分布式智能视频监控系统.本文采用基于统计学习的方法实现了运动前景图像检测和阴影消除,并在此基础上完成运动目标行为分析,实现了关键区域保护、监控目标移动检测以及镜头遮挡保护等智能视频监控功能.本文所设计的视觉分析算法适用于复杂的室外环境,适合于实时计算.在算法的实现中充分利用了该嵌入式智能图像传感器硬件设计的优点,在实验中取得了良好的效果.     

多媒体技术研究:2013——面向智能视频监控的视觉感知与处理

目的 随着视频监控技术的日益成熟和监控设备的普及,视频监控应用日益广泛,监控视频数据量呈现出爆炸性的增长,已经成为大数据时代的重要数据对象。然而由于视频数据本身的非结构化特性,使得监控视频数据的处理和分析相对困难。面对大量摄像头采集的监控视频大数据,如何有效地按照视频的内容和特性去传输、存储、分析和识别这些数据,已经成为一种迫切的需求。方法 本文面向智能视频监控中大规模视觉感知与智能处理问题,围绕监控视频编码、目标检测与跟踪、监控视频增强、视频运动与异常行为识别等4个主要研究方向,系统阐述2013年度的技术发展状况,并对未来的发展趋势进行展望。结果 中国最新制定的国家标准AVS2在对监控视频的编码效率上比最新国际标准H.265/HEVC高出一倍,标志着我国的视频编码技术和标准在视频监控领域已经实现跨越;视频运动目标检测跟踪的研究主要集中在有效特征提取和分类器训练等方面,机器学习等方法的引入,使得基于多实例学习、稀疏表示的运动目标检测跟踪成为研究的热点;监控视频质量增强主要包括去雾、去夜色、去雨雪、去模糊和超分辨率增强等多方面的内容,现有的算法均是对某类图像清晰化效果较好,而对其他类则相对较差,普适性不高;现有的智能动作分析与异常行为识别技术虽然得到了不断发展,算法的性能也在不断提高,但是从实用角度,除了简单的特定或可控场景外,还没有太多成熟的应用系统。结论 随着大数据时代的到来,智能视频监控的需求将日益迫切,面对众多挑战的同时,该研究领域将迎来前所未有的重大机遇,必将产生越来越多可以实用的研究成果。     

Generalized Mosaicing: High Dynamic Range in a Wide Field of View

We present an approach that significantly enhances the capabilities of traditional image mosaicking. The key observation is that as a camera moves, it senses each scene point multiple times. We rigidly attach to the camera an optical filter with spatially varying properties, so that multiple measurements are obtained for each scene point under different optical settings. Fusing the data captured in the multiple images yields an image mosaic that includes additional information about the scene. We refer to this approach as generalized mosaicing. In this paper we show that this approach can significantly extend the optical dynamic range of any given imaging system by exploiting vignetting effects. We derive the optimal vignetting configuration and implement it using an external filter with spatially varying transmittance. We also derive efficient scene sampling conditions as well as ways to self calibrate the vignetting effects. Maximum likelihood is used for image registration and fusion. In an experiment we mounted such a filter on a standard 8-bit video camera, to obtain an image panorama with dynamic range comparable to imaging with a 16-bit camera.     

基于深度学习的图像超分辨率复原研究进展

图像超分辨率复原（Super resolution restoration,SR）技术是图像处理领域的研究热点,在视频监控、图像处理、刑侦分析等领域具有广泛的应用需求.近年来,深度学习在多媒体处理领域迅猛发展,基于深度学习的图像超分辨率复原技术已逐渐成为主流技术.本文主要对现有基于深度学习的图像超分辨率复原工作进行综述.从网络类型、网络结构、训练方法等方面分析现有技术的优势与不足,对其发展脉络进行梳理.在此基础上,本文进一步指出了基于深度学习的图像超分辨率复原技术的未来发展方向.     

基于马尔算法的高速公路视频监控图像能见度研究

根据高速公路沿线的监控摄像机,对监控视频画面中的图像进行采集,通过对视频图像特征的分析处理,建立图像与真实场景之间的关系,根据图像特征随着真实场景的变化,运用图像处理的方法如：灰度变换、图像分割和特征提取等对图像进行图像处理,提出运用马尔算法,分别提取出目标物与背景,并将其逐一进行背景差计算,能够准确的监控图像中汽车的位置变化,确定目标物的位置,进而判别出能见度的大小。     

基于图像运动的三维重建及虚拟化应用

提出了一种稳定、快速地获取摄像机视频运动图像的三维重建方法,并对该运动图像做适当的虚拟化处理以展示重建效果。采用基于尺度不变特征点匹配的摄像机标定进行三维重建。尺度不变特征对于视频图像中的特征具有优秀敏锐的匹配能力,极大地放宽了摄像机标定对于设备上的限制,拓宽了实时三维重建的适用范围。通过对系统的一系列优化,不但提升了三维重建的精度,减少了错误匹配对摄像机标定的影响,而且进一步提升了处理速度。通过在三维重建的基础之上进行虚拟化处理,展示了本系统的三维重建效果。实验结果表明,该系统适用范围广,处理速度较快,重建精度高,实现了基于视频运动图像的三维重建。     

监控系统中运动目标检测算法分析

广泛了解当前智能监控技术已有的运动目标检测算法的基础上,对常用的运动目标几种检测算法进行了深入的分析和对比。针对相邻帧的帧间差分、背景减除法和光流场三种检测方法的不同算法的优缺点,介绍了一种基于背景差分和帧间差分相融合的多帧差分运动目标检测算法。通过对运动目标静态场景监控视频图像的准确获取,运用分析改进算法得到更为接近运动目标实际状况的影像轮廓。