NEW CORNER DETECTION ALGORITHM BASED ON MULTI-FEATURE SYNTHESIS

Corner detection is a chief step in computer vision. A new corner detection algorithm in planar curves is proposed. Firstly, from the human perception, two key characteristics are given as an amendment of the traditional corner properties. Based on the two properties, the concept of the fuzzy set is introduced into a detection. Secondly, the extracted-formulae of three groups including the features of the corner subject degree are derived. Through synthesizing the features of three groups, the judgments of the corner detection, location, and optimization are obtained. Finally, by using the algorithm the detection results of several examples and feature curves for some interested parts, as well as the detection results for the test images history in references are given. Results show that the algorithm is easily realized after adopting the fuzzy set, and the detection effect is very ideal.

GRAPH GRAMMAR METHOD FOR 3-D CORNER DETECTION

Most of local feature descriptors assume that the scene is planar. In the real scene, the captured images come from the 3-D world. 3-D corner as a novel invariant feature is important for the image matching and the object detection, while automatically discriminating 3-D corners from ordinary corners is difficult. A novel method for 3-D corner detection is proposed based on the image graph grammar, and it can detect the 3-D features of corners to some extent. Experimental results show that the method is valid and the 3-D corner is useful for image matching.

Subpixel accuracy for extracting groove center based on corner detection

Subpixel accuracy for V-groove center in robot welding is researched and a software measure to increase the accuracy of seam tracking by laser is presented. LOG （ Laplacian of Gaussian ） operator is adopted to detect image edge. Vgroove center is extracted by corner detection of extremum curvature. Subpixel position is obtained by Lagarange polynomial interpolation algorithm. Experiment results show that the method is brief and applied, and is sufficient for the real time of robot welding by laser sensors.

Robust Corner Detection Based on Multi-scale Curvature Product in B-spline Scale Space

这份报纸在 B 花键弯曲规模空间的框架论述一种多尺度的弯曲产品角落察觉技术。规模产品功能在不同规模从轮廓的弯曲产品被导出。角落被 thresholding 作为本地最大值构造越过几规模的弯曲产品结果。通过规模产品，本地化精确性和察觉表演能显著地以 CNN 标准被改进。实验也证明那个建议方法显示出坚韧性到高频率细节并且提供有希望的察觉结果。

A Parallel Algorithm for Corner Detection on Object Contour

A new parallel algorithm for corner detection on object contour is presented in the paper. In this algorithm whenever a point (pixel) is scanned, the k direction codes between the two sides of the point, which is on the edge of an object, are obtained by k-step forward and backward boundary tracking. A comer is determined by the sum of the difference between the two weighted code chains. Note that the whole chain code sequence or boundary of an object is not necessary to be extracted at all in this algorithm, and the corners are obtained immediately once the image is scanned, furthermore, what humans perceive as corners can be detected and localized by this algorithm.

Correg-Yolov3:a Method for Dense Buildings Detection in High-resolution Remote Sensing Images

The exploration of building detection plays an important role in urban planning,smart city and military.Aiming at the problem of high overlapping ratio of detection frames for dense building detection in high resolution remote sensing images,we present an effective YOLOv3 framework,corner regression-based YOLOv3(Correg-YOLOv3),to localize dense building accurately.This improved YOLOv3 algorithm establishes a vertex regression mechanism and an additional loss item about building vertex offsets relative to the center point of bounding box.By extending output dimensions,the trained model is able to output the rectangular bounding boxes and the building vertices meanwhile.Finally,we evaluate the performance of the Correg-YOLOv3 on our self-produced data set and provide a comparative analysis qualitatively and quantitatively.The experimental results achieve high performance in precision(96.45%),recall rate(95.75%),F1 score(96.10%)and average precision(98.05%),which were 2.73%,5.4%,4.1%and 4.73%higher than that of YOLOv3.Therefore,our proposed algorithm effectively tackles the problem of dense building detection in high resolution images.

Shi-Tomasi角点区域的拷贝图像检测

针对由于裁剪、翻转和旋转等产生的图像拷贝问题,提出一种Shi-Tomasi角点的拷贝检测算法.先使用Shi-Tomasi角点检测算法提取图像的局部角点;然后在以Shi-Tomasi角点为中心的圆环区域内计算特征向量的协方差描述子(多特征融合);最后通过协方差描述子的相似性度量来检测圆环区域的相似性,并以此判断检测图像是否为原图像的拷贝.实验结果证明,该检测算法对图像的裁剪、旋转等攻击具有较好的鲁棒性.

茅台成义烧房旧址钢筋混凝土行车梁结构性能分析

为了深入研究茅台成义烧房旧址行车梁的结构性能,通过现场调研和有限元模型分析,揭示了行车梁在正常使用和极端荷载作用下的性能表现,发现其存在锈蚀、裂缝和变形等问题。针对这些问题,提出了采用同形制钢材更换和碳纤维布加固等创新措施,以提高行车梁的结构稳定性和承载能力。加固措施不仅适用于茅台成义烧房旧址,也为类似工业遗产的保护与利用提供了有益参考。研究成果为工业遗产改造工程提供了科学依据,有助于推动工业遗产保护工作的深入发展。

结合小波变换的Shi-Tomasi算法遮挡图像匹配研究

目标部分遮挡或缺失是计算机视觉应用到制造业中经常遇到的问题。对现实生产中工件间遮挡的图像匹配问题进行研究,提出了小波变换和Shi-Tomasi角点检测相结合的算法。对小波分解后的低频图像进行特征点提取,舍弃图像的高频成分,有效降低噪声对图像的影响。选用常见的法兰盘作为实验对象,在多种环境下验证文中算法的可行性,最后用RANSAC算法消除误匹配。匹配结果表明,该算法不仅可以减少遮挡工件匹配的计算量,去除伪角点,加快识别速度,而且在噪声干扰、旋转、尺度变化等条件下也有较好的匹配效果。

Visual positioning of rectangular lead components based on Harris corners and Zernike moments

With the increasing necessities for reliable printed circuit board(PCB) product, there has been a considerable demand for high speed and high precision vision positioning system. To locate a rectangular lead component with high accuracy and reliability, a new visual positioning method was introduced. Considering the limitations of Ghosal sub-pixel edge detection algorithm, an improved algorithm was proposed, in which Harris corner features were used to coarsely detect the edge points and Zernike moments were adopted to accurately detect the edge points. Besides, two formulas were developed to determine the edge intersections whose sub-pixel coordinates were calculated with bilinear interpolation and conjugate gradient method. The last experimental results show that the proposed method can detect the deflection and offset, and the detection errors are less than 0.04° and 0.02 pixels.

改进的CornerNet-Saccade车辆检测算法

针对现有的CornerNet-Saccade算法在车辆检测任务中存在明显的误检和漏检现象,提出了一种改进的CornerNet-Saccade算法。首先通过加深堆叠沙漏网络结构,增强车辆高级特征提取能力;其次,增加更小尺度的attention maps以改善小目标车辆的检测能力,引入Dense Block模型和瓶颈残差单元降低堆叠沙漏网络参数的复杂度,保证层与层之间的最大信息流;最后,通过Sigmoid激活函数得到最终的检测结果。在KITTI数据库和自制数据库上对改进算法进行了仿真实验,平均精确率分别达92.56%和95.21%,检测速度分别达40 FPS和49FPS,同时在自制数据库上对原CornerNet-Saccade算法和改进算法进行了仿真实验,精确率和召回率相比原算法分别提高了3.8%和8.5%。结果表明:此改进的CornerNet-Saccade算法在车辆检测任务中具有明显优势。

基于Shi-Tomasi角点检测算法的弹道目标平动补偿

弹道目标在中段飞行的过程中,存在平动和微动两种运动方式。由于平动会破坏微多普勒结构,为了获得目标的相关参数以便对目标进行识别,需要对平动分量进行补偿。针对该问题,本文提出了一种基于Shi-Tomasi角点检测算法的弹道中段目标平动补偿方法。该算法从图像处理的角度出发,可以有效的提取出图像中的角点信息。之后,根据得到的角点信息进行回归分析,从而实现平动参数的估计。根据仿真实验结果,该算法可以实现平动分量的补偿并具有较好的鲁棒性。

Corner-Based Image Alignment using Pyramid Structure with Gradient Vector Similarity

This paper presents a corner-based image alignment algorithm based on the procedures of corner-based template matching and geometric parameter estimation. This algorithm consists of two stages: 1) training phase, and 2) matching phase. In the training phase, a corner detection algorithm is used to extract the corners. These corners are then used to build the pyramid images. In the matching phase, the corners are obtained using the same corner detection algorithm. The similarity measure is then determined by the differences of gradient vector between the corners obtained in the template image and the inspection image, respectively. A parabolic function is further applied to evaluate the geometric relationship between the template and the inspection images. Results show that the corner-based template matching outperforms the original edge-based template matching in efficiency, and both of them are robust against non-liner light changes. The accuracy and precision of the corner-based image alignment are competitive to that of edge-based image alignment under the same environment. In practice, the proposed algorithm demonstrates its precision, efficiency and robustness in image alignment for real world applications.

面向自动驾驶道路场景中异常案例的多模态数据挖掘算法

基于深度学习的视觉感知技术的发展有利于自动驾驶系统中环境感知技术的进步。然而,对于自动驾驶场景中的异常案例,目前的感知模型还存在一些问题。这是因为基于深度学习的感知模型的能力取决于训练数据集的分布。尤其是驾驶场景中的类别从未出现在训练集中,感知系统也往往很脆弱。因此识别未知类别和极端场景仍然是自动驾驶感知技术安全性的挑战。本文从处理数据集的角度出发,提出了一种新颖的多模态异常案例自动挖掘流程(corner case mining pipeline, CCMP)。为验证CCMP的有效性,在Waymo开放数据集的基础上构建了异常案例子集“Waymo-Anomaly”,该子集共有3 200个图像,每个图像都将包含文本中定义的异常案例场景。并且基于私有数据集Waymo-Anomaly,证明了CCMP针对异常案例场景挖掘的召回率可以达到91.7%。此外,还通过实验验证了目标检测器在包含异常案例的数据集中针对长尾分布的有效性。最终,希望从处理数据集的角度来提高自动驾驶感知模型在现实世界中的真实性。

基于机器视觉的芯片字符区域分割和定位算法

芯片表面的字符对分选具有重要意义,对字符的定位是分选工作的关键步骤。为了提高分选工作定位效率和定位精度,提出了一种基于改进的区域生长算法和凸包检测算法分割和定位芯片表面字符区域方法。首先,对采集图像进行预处理操作,利用改进的Canny算法获取无干扰图像边缘,将图像边缘作为区域生长法的种子点并以图像自适应阈值作为生长准则分割图像,使用最大内接矩形算法粗定位字符区域;其次,采用Harris角点检测算法获取字符角点分布位置;最后,筛选角点并提取关键点,利用一种改进的凸包检测算法定位字符区域。经过实验验证,所设计的算法能够完整的分割和定位芯片表面字符区域,定位精度和效率较原有方法提升5.3%和15.4%,满足实际工业生产的要求。

基于灰度均值的自适应FAST角点检测优化算法

光照不均、突变引起的灰度变化会影响图像特征检测效果。为此,设计一种基于灰度均值的自适应FAST-9-12角点检测算法。首先,利用特征点的延展性设计一种小面积双重检测模板,减少像素点与中心点的比较次数,提高区域正检率和检测速度;其次,依据图像局部灰度均值,在每个像素点检测模板内自适应调整阈值,避免灰度变化影响检测效果;最后,根据柔性非极大值抑制的思想设计角点半径抑制原则,筛选鲁棒性更强的角点。在Inria遥感影像数据集上的实验结果表明,FAST-9-12角点检测速度比FAST-12-16,FAST-9-16两种模板提高22%左右,因自适应阈值的提取方式不易受光照影响,检测准确率分别提高4.16和3.11个百分点,实现了图像特征快速和精准检测。

融合角点检测的点云配准研究

为提高点云配准效率,提出一种基于点云曲率信息的Harris角点提取算法,通过引入曲率约束,对传统Harris角点提取做出改进;计算每个点的曲率值并与整幅点云的平均曲率作比较,保留曲率值大于平均曲率的点云,实现对配准价值较低的点的剔除,为后续点云配准减少计算量;构建所提取角点之间的特征匹配,结合采样一致性(sample consensus initial alignment, SAC-IA)粗配准和正态分布变换(normal distributions transform, NDT)精配准,寻找最优的变换矩阵,实现源点云和目标点云的重合。通过比较不同算法在公开数据集上进行实验验证,结果表明该方法对复杂点云、多拐点点云都有较好的表现,对初始位置相差较大的点云也有良好的适用性。

一种基于边界斜率拟合的角点检测方法

图像中的角点为描述物体特征提供关键信息,是复杂应用(如图像分类、目标检测和跟踪、定位和测量)的预处理步骤,角点检测的质量将直接影响后续图像处理的有效性.在工业环境中,角点检测算法需要在各类噪声和干扰因素下,对大规模数据集进行高效处理,以实现实时和准确的角点检测.因此,研究和设计快速高效、高准确性的角点检测算法具有重要意义.针对传统算法需要进行曲率计算或曲线拟合的局限性,提出了一种基于Freeman边界链码的快速、轻量级角点检测算法,通过对Freeman链码在角点之前和之后的连续多个点进行线段的斜率拟合和夹角的阈值判定,进而识别出角点.从准确性、鲁棒性和计算速度等方面,在NRS工业图像集上与主流角点检测算法进行了对比实验.结果表明,所提出的算法具有较少漏检和误检的角点数量,并实现了更快的检测速度,在工业应用中更具有优势.

一种在尺度空间下基于边缘的角点目标检测方法

图像中的目标角点位置是实现很多计算机视觉任务的关键数据。为了克服传统检测算法产生的数据冗余问题,提出了一种在尺度空间下基于边缘的角点目标检测方法。首先,构建一个分组多层的尺度空间,将原图投影到其中后得到多个平滑图像。与此同时,采用定义的边缘算子检出平滑图像中所有边缘而得到多个按序存放的像素点集,当点集数量稳定时停止更大尺度的变换。然后,在当前尺度下,计算点集中各元素反映其角点强度的特征值。根据这些特征值变化规律检出角点的支撑集区间,并在此区间中采用高斯拟合函数确定最终的目标角点。实验表明,该方法能够检出特征显著的目标角点及其角度,其中合成图像精度在像素级,应用案例中的平均误差与图幅比约为1.5/100。

多关键点约束与深度估计辅助的单目3D目标检测算法

当前主流的单目相机3D目标检测网络采用关键点检测范式,存在关键点预测与深度估计不准确的问题,限制了单目3D检测器的性能表现.本文提出一种多关键点约束与深度估计辅助的单目3D目标检测算法Mono-Aux,利用3D检测框的角点投影点、上表面与下表面中心投影点作为3D框中心投影点的补充,通过多关键点约束提升关键点预测精度;提出一种LiDAR-Free解耦深度估计方法,在不引入激光点云数据的同时通过几何关系推导引入额外的深度估计辅助监督信号,提升深度估计的准确性.多关键点约束与深度估计辅助仅在训练阶段使用,推理阶段不引入额外的计算成本.在KITTI3D目标检测验证集和测试集上的结果显示,相较于MonoDLE基线网络,提出的MonoAux算法在目标检测精度上分别提高3.87%和4.64%,与其他SOTA方法相比,本文方法也具有显著的性能优势,甚至优于部分使用额外数据的方法.