基于CUDA的直升机旋翼桨叶摆振角快速测量方法研究

发布时间：2018-03-02 05:01

本文关键词： CUDA 双目视觉旋翼桨叶编码标记点旋翼坐标系摆振角测量　出处：《南昌航空大学》2016年硕士论文　论文类型：学位论文

【摘要】：直升机旋翼桨叶运动参数不仅是旋翼设计阶段非常重要的性能评估指标,也是直升机日常维护的重要检测项目,直接影响直升机的安全性、可靠性。针对基于立体视觉的直升机旋翼桨叶摆振角测量CPU串行算法效率不高的问题,利用图像处理单元(GPU)并行计算技术,开展了基于CUDA统一计算设备架构的直升机旋翼桨叶摆振角快速测量方法研究,主要研究内容和研究成果如下:(1)构建了基于CUDA的直升机旋翼桨叶摆振角快速测量系统,主要分为硬件和软件两个子系统。硬件子系统主要包括PC机、Cameralink摄像机、Cameralink图像采集卡、倍频器、LED光源等,根据摆振角测量系统各项技术指标要求,进行硬件功能模块设计和主要设备的选型;软件子系统主要包括旋翼图像采集、双目摄像机标定、图像预处理、基于CUDA的标记点快速检测、标记点解码匹配、摆振角计算等功能模块,构建了摆振角测量软件开发框架,编程实现了各个软件模块。(2)提出了一种基于CUDA的标记点快速检测算法。首先,对算法中最耗时的图像去噪、阈值分割、连通域标记、边缘检测等部分进行并行化设计;然后,采用多层次并行策略将大量密集运算分配到不同的图像处理单元上并行执行,利用共享内存和共享寄存器加速数据访问;最后,通过不同像素大小的旋翼桨叶图像测量实验,结果表明该方法执行效率明显高于CPU串行方法,可满足旋翼桨叶摆振角快速测量的要求。(3)提出了一种直升机旋翼桨叶摆振角测量方法。首先,采用张正友摄像机标定方法对双目摄像机进行标定;其次,采集直升机旋翼桨叶图像,对标记点进行检测、解码及匹配,得到左右图像标记点二维像素坐标;再次,计算旋翼桨叶所有标记点三维坐标;最后,在旋翼坐标系下根据旋翼桨叶摆振角角定义实现摆振角的测量。实验结果表明,标记点三维信息测量误差小于2 mm,测量的摆振角大小稳定,实现了高精度、高效的测量。
[Abstract]:Helicopter rotor blade motion parameter is not only a very important performance evaluation index in the rotor design stage, but also an important test item of helicopter daily maintenance, which directly affects the helicopter safety. Reliability. Aiming at the low efficiency of CPU serial algorithm for measuring rotor blade swing angle of helicopter based on stereo vision, the parallel computing technology of image processing unit (GPU) is used. The research on the rapid measurement method of rotor blade shimmy angle based on CUDA unified computing equipment architecture is carried out. The main research contents and results are as follows: 1) A rapid measurement system of helicopter rotor blade shimmy angle based on CUDA is constructed. The hardware subsystem mainly consists of PC camera camera camera. The software subsystem mainly includes rotor image acquisition, binocular camera calibration, image preprocessing, fast detection of mark points based on CUDA, tag point decoding and matching. In this paper, a software development framework for measuring shimmy angle is constructed, and a fast detection algorithm based on CUDA is proposed. Firstly, the most time-consuming image denoising algorithm is proposed. Threshold segmentation, connected domain marking, edge detection and other parts are designed in parallel, and then a large number of intensive operations are assigned to different image processing units for parallel execution by multi-level parallel strategy. The shared memory and register are used to accelerate the data access. Finally, the experimental results of rotor blade image measurement with different pixel sizes show that the efficiency of this method is obviously higher than that of CPU serial method. This paper presents a method to measure the swing angle of helicopter rotor blade. Firstly, the camera calibration method of Zhang Zhengyou is used to calibrate the binocular camera. Collect helicopter rotor blade image, detect, decode and match the mark points, get the 2-D pixel coordinates of left and right image marking points; third, calculate the three-dimensional coordinates of all marking points of rotor blades; finally, The measurement of the swing angle is realized in the rotor coordinate system according to the definition of the angle of the rotor blade. The experimental results show that the measuring error of the three dimensional information of the marked point is less than 2 mm, and the measured value of the swing angle is stable, which realizes the high precision and high efficiency measurement.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：V275.1

【参考文献】