Modeling Camera Image Formation Using a Feedforward Neural Network

One fundamental problem in computer vision and image processing is modeling the image formation of a camera, i.e., mapping a point in three-dimensional space to its projected position on the camera’s image plane. If the relationship between the space and the image plane is assumed to be linear, the relationship can be expressed in terms of a transfor-mation matrix and the matrix is often identified by regression. In this paper, we show that the space-to-image relation-ship in a camera can be modeled by a simple neural network. Unlike most other cases employing neural networks, the structure of the network is optimized so as for each link between neurons to have a physical meaning. This makes it possible to effectively initialize link weights and quickly train the network.

Stereoscopic Camera-Sensor Model for the Development of Highly Automated Driving Functions within a Virtual Test Environment

The need for efficient and reproducible development processes for sensor and perception systems is growing with their increased use in modern vehicles. Such processes can be achieved by using virtual test environments and virtual sensor models. In the context of this, the present paper documents the development of a sensor model for depth estimation of virtual three-dimensional scenarios. For this purpose, the geometric and algorithmic principles of stereoscopic camera systems are recreated in a virtual form. The model is implemented as a subroutine in the Epic Games Unreal Engine, which is one of the most common Game Engines. Its architecture consists of several independent procedures that enable a local depth estimation, but also a reconstruction of a whole three-dimensional scenery. In addition, a separate programme for calibrating the model is presented. In addition to the basic principles, the architecture and the implementation, this work also documents the evaluation of the model created. It is shown that the model meets specifically defined requirements for real-time capability and the accuracy of the evaluation. Thus, it is suitable for the virtual testing of common algorithms and highly automated driving functions.

Sensor planning method for visual tracking in 3D camera networks

Most sensors or cameras discussed in the sensor network community are usually 3D homogeneous, even though their2 D coverage areas in the ground plane are heterogeneous. Meanwhile, observed objects of camera networks are usually simplified as 2D points in previous literature. However in actual application scenes, not only cameras are always heterogeneous with different height and action radiuses, but also the observed objects are with 3D features（i.e., height）. This paper presents a sensor planning formulation addressing the efficiency enhancement of visual tracking in 3D heterogeneous camera networks that track and detect people traversing a region. The problem of sensor planning consists of three issues：（i） how to model the 3D heterogeneous cameras;（ii） how to rank the visibility, which ensures that the object of interest is visible in a camera＇s field of view;（iii） how to reconfigure the 3D viewing orientations of the cameras. This paper studies the geometric properties of 3D heterogeneous camera networks and addresses an evaluation formulation to rank the visibility of observed objects. Then a sensor planning method is proposed to improve the efficiency of visual tracking. Finally, the numerical results show that the proposed method can improve the tracking performance of the system compared to the conventional strategies.

Overview of Camera Calibration

This paper introduce several commonly used camera calibration method include their advantages and disadvantages.The traditional methods of camera calibration need to know precise structural information of calibration object and they are accurate.Camera calibration based on active vision get the camera calibration through the controlling of some parameters of camera.Self-calibration of camera don't need to use the calibration block and get the camera calibration through the corresponding relationship among multiple images.

Observation and theoretic analysis of gas-bubble formation and growth in water-model

The behavior of bubbles is observed with high-speed digital camera in water-model. It is found that each bubble has three processes: bubble formation, bubble coalescence and bubble division. Bubble shape is spherical firstly, then elliptical and spherical crown after coalescence, and spherical again after division. These phenomena are explained theoretically. And the bubble size is defined newly. The so-defined bubble size is measured through digital camera and LECO graphical analyzer. And the measured results are compared with those in literatures.

随机噪声平板下光学复眼内外参联合标定

光学复眼在无人系统的精确定位制导、避障导航等任务中得到了越来越广泛的应用,其中光学复眼的高精度标定是保障上述任务质量的前提。通常经典的张氏棋盘格标定法要求光学复眼的每个子眼都必须观测到完整的棋盘格,然而,由于光学复眼结构的复杂性,在实际标定过程中难以满足这一要求。为解决张氏标定法的局限性,该文提出一种基于随机噪声平板的光学复眼内外参联合标定算法,该算法通过子眼拍摄随机噪声平板的局部信息,可简单快速地实现任意构型和子眼数量的光学复眼内外参联合标定。为了提高光学复眼标定的稳定性,设置多阈值匹配机制解决子眼视场特征点数量稀疏导致图像匹配失效的问题。同时,给出了光学复眼内外参联合标定的误差模型,用来衡量所提出算法的精确度。在与张氏棋盘格标定法进行实验对比中,验证所提算法的稳定性和鲁棒性,并在光学复眼实物系统中,验证了所提联合标定算法具有较高的精度。

Using UAVs for detection of trees from digital surface models

A difficult problem in forestry is tree inventory.In this study, a GoProHero attached to a small unmanned aerial vehicle was used to capture images of a small area covered by pinus pinea trees. Then, a digital surface model was generated with image matching. The elevation model representing the terrain surface, a ‘digital terrain model’,was extracted from the digital surface model using morphological filtering. Individual trees were extracted by analyzing elevation flow on the digital elevation model because the elevation reached the highest value on the tree peaks compared to the neighborhood elevation pixels. The quality of the results was assessed by comparison with reference data for correctness of the estimated number of trees. The tree heights were calculated and evaluated with ground truth dataset. The results showed 80% correctness and 90% completeness.

车载视觉环视系统实验平台

为了帮助驾驶员更好地感知盲区和障碍物,提升驾驶的安全性与便捷性,提出了一种车载视觉环视系统实验平台。利用安装在车辆前、后、左、右的多台鱼眼相机实时采集车辆周边视频数据,生成以车辆为参考中心的环视鸟瞰图,为驾驶员提供360°无死角的环境视觉信息。最后,通过实物实验验证了所提方法的有效性。

面向车载相机采集图像的智能汽车测试场景关键性量化模型

车载相机图像是构建智能汽车测试场景库的主要数据来源,但其中关键测试场景发生概率低,大部分场景的测试价值小,若将其直接应用于智能汽车测试会浪费大量测试资源。本文提出一种面向车载相机采集图像的智能汽车测试场景关键性量化模型。首先,基于实车相机参数对实车采集的图像进行处理,输出对行车安全具有影响的参数;其次,基于风险场理论将参数整合,输出测试场景关键性量化结果;最后,对实车采集的图像进行测试场景关键性量化验证,结果表明本文模型可以精确输出测试场景关键性的具体数值,进而对比不同场景的测试价值,证明本文提出的模型可以有效筛选智能汽车关键测试场景。

基于线激光三维视觉的燃料组件水下在位变形测量方法

为了解决燃料组件弯曲与扭曲变形水下在位测量难题,提出了基于线激光三维视觉的燃料组件水下在位变形测量方法。通过基于单层折射的水下相机标定、基于平面约束的光平面标定算法研究解决单个测量单元的标定问题;研究基于三维标定工装的10组测量单元在位快速标定方法,解决测量系统全局坐标统一问题,并利用激光跟踪仪测量标定工装的弯扭精度,进行测量精度对比试验验证。研究表明,研制的基于线激光三维视觉的燃料组件水下在位变形测量系统扭曲测量误差限为0.15°,弯曲测量误差限为0.3 mm,且整个测量时间约在1 s,测量效率和测量误差限优于现有的测量系统。

基于XGBoost算法的人—虎共存区域风险等级划分

以2014—2019年珲春地区红外相机拍摄的东北虎数据为基础,基于XGBoost算法构建了虎出没区域风险等级划分模型。由模型检验可知:模型的准确率为93.51%,精确率为93.85%,召回率为93.08%,F1值为93.31%,Cohen s Kappa统计系数为90.2%。研究结果表明:基于XGBoost算法构建的人-虎共存区域风险等级划分模型分类效果好、预测准确度高,运用该模型对人-虎共存区域进行风险等级划分是可行的。

亚热带森林小麂的空间分布及活动节律的区域性差异

小麂(Muntiacus reevesi)是偶蹄目鹿科麂属动物,在我国种群数量丰富,分布广泛,但关于小麂在不同区域的空间分布及活动节律的差异性研究相对较少。为掌握小麂的活动节律及其影响因素,2018年3月至2019年2月利用红外相机技术对江西省桃红岭梅花鹿国家级自然保护区、九岭山国家级自然保护区和齐云山国家级自然保护区的小麂进行了监测,每个保护区的红外相机数量均为60台,相机工作日分别为9692、9659、14582 d,小麂的独立有效照片数分别为603、665、1032张,并运用单季节占域模型估算了各地区小麂的占域率和探测率,分析了海拔、坡度、植被覆盖度以及距水源地距离等环境变量对小麂活动规律以及分布范围的影响。结果表明,桃红岭保护区和九岭山保护区的小麂偏好于海拔较低、坡度较小、植被指数较低、距水源地距离较近的区域;齐云山保护区的小麂偏好于海拔较高、坡度较大、植被指数较高、距水源地距离较近的区域。小麂为明显的晨昏性动物,日活动高峰出现在06:00—08:00和17:00—19:00。其日活动节律在齐云山存在显著季节差异,桃红岭保护区和九岭山保护区则无显著季节差异,各保护区之间的年活动节律不存在显著差异。该研究结果有助于进一步了解小麂的活动节律模式及其季节性变化,进一步了解小麂在自然状态下的生态习性,为这一物种的保护和管理提供数据支撑。

针对十参数标定模型的导航避障相机影像精确校正

十参数标定模型是近景摄影测量中一种经典的内参标定模型。在影像畸变校正中,如果利用此模型采用直接重采样方式会产生空洞条纹,而常规的间接重采样方式虽无空洞产生,但在畸变较大的非中心区域同样无法得到较好的校正效果。针对此问题,本文提出了一种针对十参数标定模型的影像牛顿迭代精确校正算法。首先利用十参数标定模型将像点理想值和实际值进行重组,构建以像点实际值为待求量的非线性方程组;然后采用牛顿迭代法对影像中畸变部分进行局部线性逼近求解;最后采用间接重采样方法实现精确校正的目的。以移动机器人导航和避障相机作为研究对象,试验结果表明,牛顿迭代法校正可以得到高精度的优质去畸变影像,对避障相机的处理效果尤为显著。本文算法适用于移动机器人导航避障相机影像的畸变精确去除,为后续立体匹配环节生成优质的双目核线影像提供了可靠数据源。

机器人与双目相机的手眼标定方法

为了得到钻头在激光熔覆机器人坐标系下的模型数据需要机器人与双目相机进行手眼标定,针对这一问题提出了一种手眼标定的方法。改变机器人位姿对同一点进行测量,建立机器人DH模型并利用机器人正运动学方程计算出机器人末端坐标系与双目相机坐标系之间的变换矩阵,完成了手眼测量系统标定。基于此标定方法进行了误差测试与激光熔覆实验,对误差产生的原因进行分析并提出了减小误差的方法。实验结果表明,机器人与双目相机的手眼标定方法有效,可用于PDC钻头激光熔覆系统。

高密度城区野猪生境适宜性与“人猪冲突”潜在风险区识别研究

快速城市化进程引发的生境变化导致城市生物多样性整体下降,但一些能适应碎片化生境的物种却大量繁殖,被迫扩散或进入城区觅食,从而引发野生动物和居民的现实或潜在冲突。文章针对近年来城区“人猪冲突”现象,以武汉市洪山区为例,基于红外相机监测,结合MaxEnt模型,探究高密度城区野猪生境适宜性和“人猪冲突”的潜在热点区域。结果表明:洪山区野猪偏好海拔高于60 m、坡度介于10°~45°的城市林地区域,主要集中在洪山区中部的低山缓丘,共识别出潜在适宜区面积13.80 km2;洪山区潜在“人猪冲突”区域位于中部,并形成自西向东北贯穿、3个核心区聚集的格局;高密度城区野猪夜行性显著;适宜区主要分布在城市林地,不适宜区主要是建设用地;土地利用类型、高程和坡度对城区野猪种群分布影响较大;城市林地斑块及之间的建成区面临较大“人猪冲突”风险。

基于高斯增强模块的相机模型辨别

在多媒体取证中,高通滤波器是卷积神经网络常用的预处理层之一,用于抑制图像内容的影响,只强调高频特征。但与此同时,其他一些包含取证痕迹的有用信息也将被不加区别地剔除。为了解决这一问题,文中提出了一个简单而高效的高斯增强模块(Gaussian Enhancement Module,GEM)来提取“扩展的”高频特征,即在维持原有特征强度的基础上增强高频细节信息。GEM由两个连续的一维低通高斯滤波器组成,以获得一个模糊版本的特征图,并进一步得到相应的扩展高频残差。通过以高频残差作为空间掩膜,它可以自适应地强化脆弱和细微的低级取证特征,并防止在特征传递过程中出现衰减现象。在相机模型辨别数据集上进行实验,通过将该模块插入多个主流骨干网络,GEM仅仅带来非常轻微的模型复杂度的增加,网络性能和鲁棒性却显著提高,表明该模块支持“即插即用”,与特定的网络架构无关。

一种街景图像中建筑物高度估算方法

建筑物高度信息是城市三维建模的基础数据,但已有的建筑物高度估算研究多采用LiDAR和SAR等遥感影像。随着计算机和互联网的快速发展,街景数据因采集容易和成本低等特点成为了一种新兴的建筑物高度估算数据源。文章提出一种街景图像中建筑物高度估算方法,首先利用segment anything model实现图像中建筑物像素高度提取;然后利用图像元数据和电子地图数据获取建筑物与相机之间的距离、图像焦距,根据街景图像与建筑物实体的几何关系改进针孔相机模型,构建建筑物高度估算方法;最后选取北京、柏林的Mapillary街景图像开展实验验证。结果表明,与改进前相比,改进后针孔相机模型明显提升了高度估算准确度,RMSE降低了11.31 m,R^(2)提高了0.4,具备实用价值。

大视场相机最优投影模型识别及星光标定方法

针对在轨摄影测量中近距离大尺寸测量需求,提出利用星光约束的大视场角摄影测量相机最优投影模型识别及标定方法。首先,构建了具备调节系数的星光几何投影分段函数模型。随后,针对分段星光投影模型开发多站位自标定光束平差算法。通过将光束平差算法与北方苍鹰寻优策略相结合,对投影模型调节系数、相机内方位参数、相机外方位参数及镜头畸变系数同步优化,直到星点像面重投影均方根误差达到全局最小,得到最优投影模型及其参数。实测实验表明,大视场角相机星光标定后,星点像面坐标的重投影均方根误差为1/9 pixel。在连续帧星光标定实验中,通过卡尔曼滤波算法对相机参数随机误差进行了有效消除。该方法可在相机星光标定过程中识别最优投影模型并标定全部成像参数,具备连续帧标定及参数校准能力。

基于机器视觉的汽车车门三维定位引导

目前车间中常通过预先示教机械臂运动轨迹的方式对夹具上位姿固定的零部件进行引导抓取,存在人机工程性差且智能化程度较低等问题。机器视觉作为非接触感知测量的有效手段已在工业场景中有所应用。基于单目视觉位姿测量技术建立多相机n点透视定位模型和机械臂引导模型,实现了对零部件的空间感知和引导抓取。提出一种用于补偿机械臂和视觉模块系统误差的多零位锚点的预标定方法,有效减小了大尺寸零部件的定位引导误差。

基于无人机航拍的苎麻倒伏信息解译研究

茎杆倒伏是苎麻三麻培育中最常见的灾害,传统的监测方法具有耗时耗力、不及时等局限性。提出了一种基于无人机航拍获取苎麻倒伏信息的方法,首先利用Pix4D Mapper软件生成苎麻的冠层正射影像和数字表面模型(digital surface model,DSM),基于正射影像提取苎麻光谱、纹理及形状特征,基于DSM提取苎麻株高指标,最后结合3种机器学习算法构建正常/倒伏苎麻分类模型。结果表明,基于DSM提取的株高信息可以有效代替大田实测株高,模型R2为0.899。倒伏和正常苎麻在光谱、纹理、形状及株高特征上具有差异。在3种机器学习算法中,支持向量机和决策树模型的性能最好,准确率达到99%,能够高效地识别苎麻倒伏地块。以上研究结果为准确、快速评估作物倒伏情况提供了技术支撑。