移动机器人视觉中图像特征点提取与匹配技术研究

发布时间：2018-02-24 12:13

本文关键词： 视觉导航三维重建高纹理低纹理超像素匹配　出处：《浙江理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：移动机器人视觉导航中,为了完成机器人的定位和导航,就要通过视觉系统获取环境的二维图像信息,进而完成三维重建。图像匹配是整个三维重建中最为重要的过程。本研究根据环境的不同纹理信息,制定了对应的研究方案和流程。首先对高纹理图像进行了分析,成功完成了图像间的匹配。但是匹配的前提是图像像素具有高梯度特征。对于环境中包括低纹理(如平坦的桌面)区域的情况下,由于图像像素的梯度很低,导致单纯的使用上述方案进行匹配很难在重建中得到质量高的三维效果。因此在图像的低纹理区域先使用图像分割技术得到超像素,之后超像素的匹配运用迭代的思想和数理统计的方法,通过迭代搜索,筛选满足条件的匹配对象。那么就可以把这个方案运用到机器人视觉导航中来,从而完成环境的三维重建。通过实验表明,在低纹理图像区域使用超像素匹配,提高了三维重建的精度。为了很好的完成环境的三维重建,首先需要对获取环境的相机进行标定,得到相机的参数。这个参数是二维矩阵图像坐标和三维世界坐标之间桥梁的枢纽。之后先完成高纹理匹配,再完成低纹理匹配,最后完成了环境的重建。实验阶段,通过选择不同环境中重建的效果对比分析,得到了各种匹配方案的优缺点。整个的研究方案在实验过程中也表现出了其缺陷,不稳定性也是今后研究的方向。本研究在阅读了大量的参考文献之后,分析问题的瓶颈。通过对传统图像特征匹配方法进行分析,掌握了方法中使用到的理论,并且完成了分步过程的实验,最后选择不同纹理的图像,使用该方法分别完成了匹配实验,得知方法在低纹理图像中完成匹配的缺陷。在此基础上总结前人的经验和成果,运用最新的研究成果,完成课题目标。整个的研究中运用C++语言,OpenCV,PCL等库完成了三维建模,通过重建结果的分析对比得知,本文使用的方法解决了三维重建和视觉导航过程中的低纹理环境难匹配的问题。
[Abstract]:In the visual navigation of mobile robot, in order to complete the localization and navigation of the robot, it is necessary to obtain the two-dimensional image information of the environment through the visual system. Image matching is the most important process in the whole 3D reconstruction. According to the different texture information of the environment, the corresponding research scheme and process are worked out in this study. Firstly, the high texture image is analyzed. The matching between images has been successfully completed. But the premise of the matching is that the image pixels have a high gradient feature. In the case of an environment with a low texture (such as a flat desktop) region, the gradient of the image pixel is very low. As a result, it is difficult to get high quality 3D effect in reconstruction by using the above scheme. Therefore, in the low texture region of the image, the image segmentation technique is first used to obtain the super-pixel. Then the super-pixel matching uses iterative thought and mathematical statistics method, through iterative search, the matching objects that meet the conditions can be selected. Then this scheme can be applied to robot vision navigation. Experimental results show that the accuracy of 3D reconstruction is improved by using super-pixel matching in low-texture image area. First of all, we need to calibrate the camera to get the camera parameters. This parameter is the bridge between the two-dimensional matrix image coordinate and the three-dimensional world coordinate, and then completes the high texture matching, then completes the low texture matching. Finally, the reconstruction of the environment is completed. In the experiment stage, the advantages and disadvantages of various matching schemes are obtained by comparing and analyzing the effects of the reconstruction in different environments. Instability is also the research direction in the future. After reading a large number of references, this study analyzes the bottleneck of the problem. By analyzing the traditional image feature matching method, we have mastered the theory used in the method. Finally, we select different texture images, and use this method to complete matching experiments, and find out the defects of the method in low texture images. On this basis, we summarize the previous experience and results. Using the latest research results to achieve the goal of the project. In the whole study, we have completed the 3D modeling by using C language OpenCVC PCL and other libraries. Through the analysis and comparison of the reconstruction results, we know that, The method used in this paper solves the problem of low texture environment matching in 3D reconstruction and visual navigation.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【相似文献】