面向位姿估计的相机系统标定方法研究

发布时间：2019-03-16 10:43

【摘要】：近年来,机器人在工业生产线上的应用日益得到密切关注。在柔性生产线上,根据客户需求,快速适应不同工业品的生产,对机器人系统提出了挑战。计算机视觉与机器人技术的结合可以大大提高工业生产线的适应能力。通过双目视觉系统赋予机器人立体视觉能力,是提高机器人抓取适应性的有效途径。在该系统中,研究具有高精度的相机标定和目标位姿估计方法是两个关键问题。本论文主要创新点和研究成果如下：镜头畸变是影响成像品质的主要因素,为了提高畸变矫正精度,提出了灭点重投影模型,利用该模型能有效地改善在畸变图像中对直线方程估计的精度。在非线性优化过程中,针对参数耦合导致优化结果落入局部最优的问题,提出了依次迭代参数优化方法。在此基础上,提出了一种基于灭点重投影的畸变矫正算法。通过仿真实验评估噪声对矫正精度的影响,实验结果表明无噪声条件下的矫正精度为1.5像素,随着噪声强度增加误差缓慢增大,说明该算法有一定的抗噪声能力；在实验室环境下,实验测试表明,该方法能有效地矫正畸变图像。相机标定是双目视觉系统获取空间坐标的关键步骤,为了提高相机标定精度,提出了一种基于灭点一致性约束模型的相机标定算法。该模型将灭点重投影畸变矫正方法与单应矩阵求解方法相结合,通过提高灭点位置精度达到提高畸变矫正精度的目的,最终提高相机标定精度。仿真实验结果表明,平均重投影误差为0.04像素,能从具有较大畸变的图像中获得较高的重投影精度；在实验室环境下,实验测试结果表明,该算法最大重投影误差分别为0.60像素与0.50像素。从目标坐标数据中计算出目标位姿是实现机器人抓取的重要步骤,为了提高位姿估计算法的适应性,提出了一种最远点剔除的改进位姿估计算法。该算法针对传统迭代最近点(ICP)方法对初始位姿敏感的问题,引入了混合单纯形模拟退火算法,在参数空间中搜索最优解,减小了初始位姿对结果的影响。为了排除误匹配点的干扰,提出了一种最远点剔除策略。仿真实验结果表明,改进的算法优于传统ICP方法。设计了双目立体视觉实验系统,提出并设计了一种基于DSP/BIOS的多任务软件架构,成功地实现了目标位姿估计。在实验系统中,利用KUKA机器人控制目标位姿变化,以评估位姿估计精度,实验结果表明,该方法的平移总误差为2.7mm,满足目标抓取的需求。
[Abstract]:In recent years, the application of robot in industrial production line has been paid more and more attention. On the flexible production line, the robot system is challenged to adapt to the production of different industrial products quickly according to the customer's demand. The combination of computer vision and robot technology can greatly improve the adaptability of industrial production line. It is an effective way to improve the robot's grasping adaptability by giving robot stereoscopic vision ability through binocular vision system. In this system, it is two key problems to study camera calibration and target pose estimation with high precision. The main innovations and research results of this thesis are as follows: lens distortion is the main factor affecting the imaging quality. In order to improve the distortion correction accuracy, the double projection model of extinction point is proposed. This model can effectively improve the accuracy of linear equation estimation in distorted images. In the process of nonlinear optimization, a sequential iterative parameter optimization method is proposed to solve the problem that parameter coupling causes the optimization results to fall into local optimization. On this basis, a distortion correction algorithm based on re-projection of vanishing point is proposed. The effect of noise on the correction accuracy is evaluated by simulation experiments. The experimental results show that the correction accuracy is 1.5 pixels under the condition of no noise, and the error increases slowly with the increase of noise intensity, which shows that the algorithm has a certain anti-noise capability. In the laboratory environment, the experimental results show that the method can effectively correct distorted images. Camera calibration is a key step for binocular vision system to obtain spatial coordinates. In order to improve camera calibration accuracy, a camera calibration algorithm based on the consistency constraint model of vanishing point is proposed. The model combines the double projection distortion correction method with the unitary matrix method to improve the accuracy of distortion correction and finally improve the calibration accuracy of the camera by improving the position accuracy of the vanishing point. The simulation results show that the average re-projection error is 0.04 pixels, and the precision of re-projection can be obtained from the image with large distortion. The experimental results show that the maximum double projection error of the algorithm is 0.60 pixels and 0.50 pixels, respectively. It is an important step to calculate the position and pose of the robot from the target coordinate data. In order to improve the adaptability of the pose estimation algorithm, an improved position and pose estimation algorithm with the farthest point elimination is proposed. In order to solve the problem that the traditional iterative nearest point (ICP) method is sensitive to the initial position and pose, a hybrid simplex simulated annealing algorithm is introduced to search the optimal solution in the parameter space, thus reducing the influence of the initial position on the result. In order to eliminate the interference of mismatched points, a farthest point elimination strategy is proposed. The simulation results show that the improved algorithm is better than the traditional ICP method. A binocular stereo vision experimental system is designed and a multi-task software architecture based on DSP/BIOS is proposed and designed. The target pose estimation is successfully realized. In the experimental system, the KUKA robot is used to control the position and pose change of the target to evaluate the accuracy of the position and pose estimation. The experimental results show that the total translation error of the method is 2.7 mm, which meets the requirements of target grabbing.
【学位授予单位】：华中科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】