图像的线特征提取算法研究及ARM实现

发布时间：2018-12-21 21:35

【摘要】：在一幅图像当中,图像的线特征(包括直线、曲线)往往组成了物体的轮廓,如果能够有效地将这些特征提取出来,则对于图像处理在目标匹配、目标识别、机器人视觉、数据挖掘等领域的应用具有重要意义。本文针对数字图像中存在不同特点的线特征提取进行了算法研究,主要研究内容包括:首先,针对图像中存在具有非平滑边缘的曲线特征(圆或椭圆)的情况进行了特征提取算法的研究,提出一种基于多尺度边缘提取的圆及椭圆检测算法,逐一检测利用高斯金字塔所得到的不同尺度图像,以此提高曲线特征的检测率,并将该算法应用于天体表面陨石坑图像的检测,并且检测率可以达到85%以上,最后将该算法与Fornaciari等人提出的算法(Forn算法)进行检测结果比较,结果表明,与Forn算法相比,检测率提高40%,检测时间提高50%以上。然后,针对图像中存在多种线特征并存的情况,研究了一种基于ELSD的直线与椭圆提取算法。首先利用区域生长以及曲线生长的方法分别对直线段和椭圆弧段进行检测,通过分析椭圆弧段的位置信息和切线方向信息进行最小二乘算法的椭圆拟合,并利用计算误报率来剔除拟合出来的无效椭圆。应用该算法对道路交通标志图像进行检测,实验结果表明,该算法不仅能够对存在多种线特征的交通标志图像有较好的检测结果,且检测快速。最后,针对基于ARM平台的圆(椭圆)检测系统进行了设计和研究。通过对硬件部分和软件部分的设计实现该系统的整体搭建。在硬件部分中,以i.MX6DL处理器为核心,并结合一系列外围接口;软件部分则利用OpenCV(跨平台计算机视觉库)以及QT等软件资源来实现嵌入式系统下的图像的处理及显示。最后将交通标志检测程序移植到该系统中,实现了基于ARM平台的交通标志检测。
[Abstract]:In an image, the line features (including lines, curves) of an image often form the contour of an object. If these features can be effectively extracted, then for image processing in object matching, target recognition, robot vision, The application of data mining and other fields is of great significance. In this paper, the algorithm of line feature extraction with different characteristics in digital image is studied. The main research contents are as follows: first of all, In this paper, the feature extraction algorithm is studied for the existence of curve features (circles or ellipses) with non-smooth edges, and a multi-scale edge detection algorithm based on circle and ellipse is proposed. In order to improve the detection rate of curve features, the algorithm is applied to the detection of crater images on the surface of celestial bodies one by using the different scale images obtained from Gao Si's pyramids, and the detection rate can reach more than 85%. Finally, the detection results of this algorithm are compared with the Forn algorithm proposed by Fornaciari et al. The results show that compared with the Forn algorithm, the detection rate is increased by 40% and the detection time is increased by more than 50%. Then, an algorithm of line and ellipse extraction based on ELSD is proposed. Firstly, the linear segment and the elliptical arc segment are detected by the method of region growth and curve growth, and the elliptical fitting of the elliptical arc segment and tangent direction information is carried out by analyzing the position information of the elliptic arc segment and the tangent direction information. The invalid ellipse is eliminated by calculating false alarm rate. The experimental results show that the algorithm can not only detect the traffic sign images with multiple line features, but also can detect the traffic signs quickly. Finally, the circle (ellipse) detection system based on ARM platform is designed and studied. The whole system is built by the design of hardware and software. In the hardware part, the i.MX6DL processor is used as the core and a series of peripheral interfaces are combined, while the software part uses the software resources such as OpenCV (cross-platform computer vision library) and QT to realize the image processing and display in the embedded system. Finally, the traffic sign detection program is transplanted into the system, and the traffic sign detection based on ARM platform is realized.
【学位授予单位】：青岛科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】