玉米叶片点云去噪软件的设计与实现

发布时间：2018-06-15 21:34

本文选题：点云去噪 + 参数　；参考：《西北农林科技大学》2017年硕士论文

【摘要】：作物模型三维重建以及可视化是当前国内外农业信息化研究的重点领域,为了获得高质量三维点云数据,点云去噪成为热门的研究领域。目前,点云去噪方面的算法都是将图像去噪方法移植到点云模型上,而单一的某种算法难以满足多种形态点云的去噪要求。基于此,本文针对玉米叶片,研究其去噪问题,并结合PC端与移动端设计并开发了一款玉米叶片点云去噪软件。本文主要的研究内容如下:(1)提出了一种移动端和PC端结合的点云去噪方法。针对点云数量过多,离群点太多,导致运行效率慢,无法直接在移动端进行去噪等问题,分析点云精简算法和去噪算法,结合玉米叶片和平台的特点,在点云精简方面采用体素化网格下采样简化算法;PC端去噪采用K近邻点距离统计去噪算法,该方法可以通过多次调整K邻近点的个数和阈值来达到去除明显噪声点的效果。移动端去噪采用双边滤波去噪算法,该方法可以在保持叶片边缘特征的情况下去除小尺度噪声点。(2)设计和开发点云去噪软件。针对去噪和精简算法的参数选择范围太广,普通用户无法选择合适参数的问题,本文使用数量不同的点云数据进行去噪实验,结果表明阈值的选择对点云去噪有较大的影响,阈值选择过大,达不到去噪的效果,选择过小,会导致去噪程度过大出现大量的孔洞;而邻近点数量在阈值选取合理范围内会起到保护叶片形态的作用。当玉米叶片为80000左右的时候,体素化网格边长选择1.2～2.6之间,玉米叶片点云精简率为30%～60%,此时阈值选择1.2～1.6之间,邻近点数量为80时在PC端去除离群点的效果较好,然后在移动端进行小尺度噪声去除,观察法向量方向,去噪处理后,杂乱的法向量变得有序,去噪时间在15.3s以内,在效果和时间上均得到较为理想的结果。(3)软件测试。针对软件存在的潜在问题,编写了较为详细的测试用例,对软件进行数据输入测试和功能测试,测试结果表明软件在输入不合理的或者非法的数值不会使程序中止,而输入合理范围内的数值会得到相应的结果,对超出程序范围的输入给出提示,不同形态、不同数量的玉米叶片点云去噪效果均可满足要求,在移动端测试时,数据量不合理会给出相应的提示。
[Abstract]:Three-dimensional reconstruction and visualization of crop models are the key areas of agricultural information research at home and abroad. In order to obtain high-quality 3D point cloud data point cloud denoising has become a hot research field. At present, the image denoising method is transplanted to the point cloud model, but a single algorithm is difficult to meet the needs of a variety of morphological point cloud de-noising. Based on this, this paper studies the problem of corn leaf de-noising, and designs and develops a corn leaf point cloud denoising software combined with PC and mobile side. The main contents of this paper are as follows: 1) A point cloud denoising method combining mobile and PC is proposed. In view of the problems of too many point clouds and too many outliers, which result in slow operation efficiency and the inability to carry out de-noising directly at the mobile end, this paper analyzes the point cloud reduction algorithm and denoising algorithm, combining with the characteristics of maize leaves and platforms. In the aspect of point cloud reduction, a voxel mesh sampling simplification algorithm is used to de-noise the PC end. The K-nearest neighbor point distance statistical de-noising algorithm is adopted. This method can remove obvious noise points by adjusting the number and threshold of K adjacent points several times. The two-sided filter denoising algorithm is used in the mobile end denoising. This method can remove the small scale noise point while keeping the edge feature of the blade. (2) the point cloud denoising software is designed and developed. Aiming at the problem that the parameter selection range of the denoising and reduction algorithms is too wide and ordinary users can not choose the appropriate parameters, this paper uses different number of point cloud data to carry out the denoising experiment. The results show that the choice of threshold has a great influence on the point cloud denoising. If the threshold value is too large, the effect of de-noising can not be achieved, and too small a choice will lead to a large number of holes in the denoising degree, and the number of adjacent points will protect the leaf morphology within a reasonable range of threshold selection. When the maize leaves were about 80000, the volumetric grid length was 1.22.6.The point cloud reduction rate of maize leaves was 300.600.When the threshold value was between 1.2 and 1.6, and the number of adjacent points was 80, the effect of removing outliers at the PC end was better when the threshold value was between 1.2 and 1.6, and when the number of adjacent points was 80, it was better to remove outliers at the PC end. Then the small scale noise removal is carried out at the mobile end, the direction of the normal vector is observed. After denoising, the chaotic normal vector becomes orderly, the denoising time is less than 15.3s, and the result of software testing is satisfactory in both effect and time. Aiming at the potential problems of the software, a more detailed test case is written to test the data input and function of the software. The test results show that the software does not suspend the program in the input of unreasonable or illegal values. And the values within a reasonable range of input will get the corresponding results, the input beyond the scope of the program to give a hint, different shapes, different quantities of corn leaf point cloud denoising effect can meet the requirements, in the mobile side test, Unreasonable amount of data will give a corresponding hint.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S513;TP391.41

【参考文献】