彩色图像对自适应匹配算法

提出了两种方法来提高图像匹配的精度。一种方法是改变自适应匹配窗口的选择方法,与原来的矩形窗口相比较,提出的窗口选取方法能在低纹理区域得到包含有效信息量更多、更接近实际边界的窗口。此外,通过降低边界点相似度在计算过程中的权重来降低边界点影响力,使得视差不连续区域的匹配精度得到提高。算法采用了Middlebury网站上提供的四幅立体图像对Tsukuba、Venus、Teddy和Cones分别进行实验验证。实验表明文中提出的两种方法对四幅图像的匹配精度均有所提高。     

利用双目视差理论的摄像机参数标定方法

针对龙眼和荔枝采摘机械手的目标获取问题,文章提出了基于双目立体视觉的摄像机标定的实验系统。实验系统是采用双目视差原理来获取深度信息的标定模型,通过对不同距离的标定板上已知特征点的图片获取和数据处理和分析,确定了两摄像机之间的最佳基线距离,在此基础上进一步验证了采摘目标的深度信息与摄像机的焦距有关的理论,并获得摄像机的合理焦距,为采摘机械手定位提供了基础理论研究依据。     

双目PMP视觉测量中立体匹配方法的设计与实现

针对PMP视觉测量中物体深度变化大或表面形状复杂时出现的误匹配问题,提出一种基于区域分割和视差修正的立体匹配方法。研究了四步相移法和多频外差相移的基本原理,并采用基于OpenCV的标定立体校正方法对光栅图像进行了校正;结合PMP视觉测量系统的特点对相机视场进行分析,根据视场各区域的相位信息特点以及视差约束,提出基于混合阈值的区域分割原则和实现方法;应用视差求解和视差校正获得有效的视差图,并在自主研发的PMP视觉测量系统中验证了立体匹配方法的有效性。     

On the smoothness constraint in the intensity-based estimation of the parallax field

In this contribution, the estimation of the parallax field is considered in a mathematical way. First, a mathematical model for the image formation is developed : This model describes how the parallax field influences the left and the right image. Then, we tackle the inverse problem : We calculate the parallax field given the left and the right image. In fact, the problem of estimating the parallax field is mathematically formulated as the maximum a posteriori probability (MAP) estimation of a Gaussian signal given two (noisy) measured signals, i.e. the left and the right image. Estimation problems like this are considered in [2]. We apply that theory to the problem at hand and prove that the MAP estimate for the parallax field minimizes a certain functional. This functional is compared to the functional of the traditional intensity-based method. Both functionals consist of a term associated with the displaced frame difference and a term associated with the smoothness of the parallax field. Our functional differs from the traditional one in the smoothness term. Next, we consider the searching for the minimum of this functional and develop a new iterative procedure to obtain the MAP estimate in a few iterations. Our procedure is quite different from the Gauss-Seidel or Jacobi schemes [3] that are normally used for this purpose. Finally, some results are given to compare the performance of our smoothness constraint to the traditional one.     

视差和运动对立体视频舒适度感知的影响

为了更好地评价立体视频的舒适程度,提出了将立 体视频中存在的3种视差与物体的运动特性作为影响立体视频舒 适度重要因素,并利用这4种因素在不同的立体视频内容中得到的数据变化建立符合人眼感 知的客观舒 适度评价标准。首先提取帧内显著物体视差、帧内相关背景视差、帧间视差和运动变量4种 因素;然后结合深度感知理论和最小二乘拟和方法建立评价模型;最后 利用主观 评价的方法对模型进行评定。结果表明,本文方法可以很好地模拟人眼的 主观感受。     

Performance improvement of multi‐view video coding based on geometric prediction and human visual system

This article proposed an accurate disparity vector prediction (DVP) algorithm for multi‐view video coding. Differing from traditional DVP that uses the information of motion vectors of neighboring blocks, the geometry of the camera position is utilized to calculate the parallax of different viewpoints in this algorithm and this parallax is the foundation of DVP. We jointly applied the Just‐Noticeable‐Difference human visual model to the DVP. After filtered using Gaussian function, the geometric DVP was obtained. Experimental results showed that the proposed method achieved significant data reduction and subjective/objective quality enhancement. © 2015 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 25, 41–49, 2015     

三维形态投影图视错觉的消除与利用

针对观看体视投影图出现体视合像变形、非单义性轴测投影图的视错觉,运用多学科的知识来分析原因,探讨消除对策,得到了一些重要的结论．是拓宽和深化体视投影理论、轴测投影理论,提高其运应用质量的一项研究     

Reflective and transflective color LCDs with double polarizers

Abstract— Reflective color LCDs with double polarizers have been developed by means of optimizing liquid-crystal modes, aperture ratio, color-filter properties, thickness of the glass substrate, polarizers, reflectors, etc. These LCDs are sufficiently bright, display many colors, have good hue, and are light weight with thin outline and low power consumption. Since they have double polarizers, a very high contrast ratio can be obtained. Therefore, they are superior in text character displays. Moreover, transflective color LCDs have been developed without visible deterioration of reflective displays by adoption of a reflective polarizer and backlight system in place of a lower polarizer and reflector.     

双目机器视觉及RetinaNet模型的路侧行人感知定位

目的 行人感知是自动驾驶中必不可少的一项内容,是行车安全的保障。传统激光雷达和单目视觉组合的行人感知模式,设备硬件成本高且多源数据匹配易导致误差产生。对此,本文结合双目机器视觉技术与深度学习图像识别技术,实现对公共路权环境下路侧行人的自动感知与精准定位。方法 利用双目道路智能感知系统采集道路前景图像构建4种交通环境下的行人识别模型训练库;采用RetinaNet深度学习模型进行目标行人自动识别;通过半全局块匹配（semi-global block matching,SGBM）算法实现行人道路前景图像对的视差值计算;通过计算得出的视差图分别统计U-V方向的视差值,提出结合行人识别模型和U-V视差的测距算法,实现目标行人的坐标定位。结果 实验统计2.5 km连续测试路段的行人识别结果,对比人工统计结果,本文算法的召回率为96.27%。与YOLOv3（you only look once）和Tiny-YOLOv3方法在4种交通路况下进行比较,平均F值为96.42%,比YOLOv3和Tiny-YOLOv3分别提高0.9%和3.03%;同时,实验利用标定块在室内分别拍摄3 m、4 m和5 m不同距离的20对双目图像,验证测距算法,计算标准偏差皆小于0.01。结论 本文提出的结合RetinaNet目标识别模型与改进U-V视差算法能够实现对道路行人的检测,可以为自动驾驶的安全保障提供技术支持,具有一定的应用价值。