结合学习特征的图像矩视觉伺服方法

发布时间：2018-04-21 23:29

本文选题：视觉伺服 + 图像矩　；参考：《华南理工大学学报(自然科学版)》2017年02期

【摘要】：针对平面图像特征选择,提出一种结合学习特征的改进图像矩视觉伺服方法,以解决不变图像矩特征存在的交互矩阵奇异问题并获得更优的运动特性.该方法首先基于不变矩特征具有的TRS(2D平移、2D旋转及尺度变化)不变特性,利用非线性支持向量机回归算法,对不变矩特征与摄像机X轴、Y轴转角的关系进行分别学习建模;然后利用两个模型的估计值(即学习特征)作为针对X轴旋转及Y轴旋转运动的图像特征,其交互矩阵具有完全解耦及线性特性,且对于任意平面目标不存在奇异问题;进而结合目标图像重心点坐标和面积的归一化特征及图像方向角特征,实现摄像机在任务空间的平移及旋转六自由度控制;最后通过仿真验证了文中方法的有效性.
[Abstract]:An improved image moment visual servo method based on learning features is proposed to solve the interaction matrix singularity problem of invariant image moment features and obtain better motion characteristics. Firstly, based on the TRS(2D translation 2D rotation and scale variation of the moment invariant feature, the nonlinear support vector machine regression algorithm is used to model the relationship between the moment invariant feature and the camera X axis Y axis rotation angle. Then, the estimated values (learning characteristics) of the two models are used as the image features for X-axis rotation and Y-axis rotation. The interaction matrix is completely decoupled and linear, and there is no singular problem for any planar target. Combined with the normalized feature of the center of gravity point coordinate and area of the target image and the direction angle feature of the image, the translation and rotation of the camera in the task space can be controlled with six degrees of freedom. Finally, the effectiveness of the proposed method is verified by simulation.
【作者单位】：华南理工大学机械与汽车工程学院;
【基金】：国家高技术研究发展计划(863计划)项目(2015AA043005) 广东省数控一代机械产品创新应用示范工程专项资金资助项目(2013B011301026)~~
【分类号】：TP391.41

【相似文献】