基于正交双目视觉的菜粉蝶3D运动信息获取技术研究

发布时间：2018-03-09 10:16

本文选题：菜粉蝶　切入点：运动轨迹　出处：《西北农林科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着农业的快速发展,昆虫行为分析的重要性愈发凸显。随着对昆虫行为分析方法研究的逐渐深入,计算机视觉技术已经成为此方面研究的重要方法。本研究针对昆虫行为学研究定量、精确地获取昆虫三维运动轨迹及运动参数的需要,根据风洞实验环境和菜粉蝶运动特征,以农业常见害虫菜粉蝶为研究对象,设计了顶视和侧视摄像机正交配置的双目视觉测量系统,利用KCF(Kernelized Correlation Filters)算法,提出了一种昆虫三维轨迹及飞行参数获取方法。本研究可以较为准确地获取昆虫飞行轨迹及运动参数,可以满足对昆虫行为分析的应用要求。论文的研究内容和取得的成果如下:(1)设计了一种正交双目视觉昆虫视频采集系统。通过分析风洞的安装位置、尺寸规格和昆虫行为实验室的实际空间环境,以及获取高质量菜粉蝶活动视频数据的应用需求,选择了适于本研究的摄像机型号,设计了正交双目视觉支架和摄像机安装托板,通过分析及试验确定并设计了同步触发方式,对计算机硬件升级。测试结果表明,实际帧数可以达到理论帧数的99.93%~100%,该系统能够满足对菜粉蝶飞行视频的流畅、无丢帧的采集需求。(2)对比分析了正交双目视觉与传统平行双目视觉的摄像机标定方法。平行双目视觉需要利用2个摄像机的内外参数进行双目校正与视差图的获取,正交双目视觉没有复杂的双目校正过程,只需要通过摄像机标定得到2个摄像机各自的内部参数即可。根据研究需求以及试验的场地等外界因素,选择张正友的棋盘格标定法,对2个摄像机进行标定得到内部参数,结果表明,俯视摄像机和顶视摄像机的标定像素误差均小于0.52,可以满足标定需求。(3)研究了菜粉蝶目标的跟踪方法,得出KCF算法的最佳参数选择为线性核函数和灰度特征。根据菜粉蝶图像序列的特点和菜粉蝶本身的飞行特性,选取3种非常具有代表性的视觉目标跟踪方法进行分析,并对3种算法进行跟踪试验,对比分析表明,STC(Spatio Temporal Context Learning)算法和KCF算法跟踪丢帧率均为0,平均像素误差分别为12.47和7.12,后者更适用于菜粉蝶目标识别与跟踪。设计KCF跟踪算法不同核函数和特征试验,结果表明,采用线性核函数和灰度特征,KCF跟踪算法的平均像素误差为5.58,能有效地跟踪菜粉蝶目标。(4)研究分析了正交双目视觉的立体匹配方法。通过与传统的平行双目视觉进行对比,在保证正交性的基础上,正交匹配利用2个摄像机拍摄平面的正交关系可直接得到菜粉蝶目标的三维空间坐标,有效地降低了立体匹配算法复杂度,大大提高了获取运动目标的空间三维轨迹的效率,且跟踪丢帧率为0,能够满足本研究的系统要求。测试结果表明,本文方法可对菜粉蝶目标进行跟踪、获取其飞行3维运动轨迹,并可方便地测定菜粉蝶飞行的速度、加速度等参数,检测的三维空间对象距离相对误差小于3.44%。
[Abstract]:With the rapid development of agriculture, the importance of insect behavior analysis becomes more and more important. Computer vision technology has become an important research method in this field. In this study, according to the experimental environment of wind tunnel and the movement characteristics of rapeseed butterflies, according to the needs of quantitative analysis of insect behavior research and accurate acquisition of three-dimensional motion trajectories and parameters of insects, In this paper, the binocular vision measurement system with the orthogonal configuration of top and side view cameras is designed, and the algorithm of KCF(Kernelized Correlation filters is used to design a binocular vision measurement system, which is a common agricultural pest, rapeseed butterfly. In this paper, a method for obtaining flight parameters and flight trajectory of insects is proposed, and the flight trajectory and motion parameters of insects can be obtained more accurately in this study. The research contents and results of this paper are as follows: 1) an orthogonal binocular visual insect video acquisition system is designed. By analyzing the installation position of wind tunnel, The dimension specification and the actual space environment of insect behavior laboratory, as well as the application requirement of obtaining high-quality video data of rapeseed butterfly activity, selected the camera model suitable for this study, designed the orthogonal binocular visual support and camera mounting plate. Through analysis and experiment, the synchronous trigger mode is determined and designed, and the computer hardware is upgraded. The test results show that the actual frame number can reach 99.933% of the theoretical frame count, and the system can satisfy the fluency of flying video of rapeseed butterfly. The camera calibration method of orthogonal binocular vision and traditional parallel binocular vision is compared and analyzed. Parallel binocular vision needs to use the internal and external parameters of two cameras for binocular correction and parallax image acquisition. Orthogonal binocular vision has no complicated binocular correction process. It only needs to get the internal parameters of two cameras by camera calibration. According to the external factors such as research demand and experimental site, Zhang Zhengyou's checkerboard calibration method is selected. The internal parameters are obtained by calibrating the two cameras. The results show that the calibration pixel errors of both overhead and overhead cameras are less than 0.52, which can meet the calibration requirements. (3) the tracking method of rapeseed butterfly target is studied. The optimal parameters of KCF algorithm are linear kernel function and gray feature. According to the characteristics of image sequence and flight characteristics of rapeseed butterfly, three very representative visual target tracking methods are selected to analyze. And the tracking experiments of three algorithms are carried out. The comparative analysis shows that the tracking frame loss rate of Spatio Temporal Context learning) algorithm and KCF algorithm are both 0, and the average pixel error is 12.47 and 7.12 respectively. The latter is more suitable for target recognition and tracking of rapeseed butterfly. Different kernel functions and feature tests of KCF tracking algorithm are designed, and the results show that, Using linear kernel function and gray feature KCF tracking algorithm, the average pixel error is 5.58, which can effectively track the target of rapeseed butterfly. The stereo matching method of orthogonal binocular vision is studied and analyzed. The method is compared with the traditional parallel binocular vision. Based on the guarantee of orthogonality, orthogonal matching can directly obtain the three-dimensional coordinates of the target of rapeseed butterfly by using the orthogonal relation of two cameras, which effectively reduces the complexity of stereo matching algorithm. The efficiency of acquiring 3D trajectory of moving target is greatly improved, and the tracking frame loss rate is 0, which can meet the system requirements of this study. The test results show that the method can track the target of rapeseed butterfly. The 3D motion trajectory is obtained, and the parameters such as velocity and acceleration of rapeseed butterfly flying can be measured conveniently. The relative error of the measured 3D object distance is less than 3.44.
【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】：S433;TP391.41

【相似文献】