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．

Proposed in this paper is an improved image moment-based visual servoing method with learning features for planar target,which helps to overcome the singularity of interaction matrix for classical invariant moment features.In the investigation,first,based on the TRS ( 2D translation,2D rotation and scale transformation) -invariant properties of invariant moment features,the nonlinear support vector machine regression algorithm is used to reveal and model the relationship between a set of specific invariant moment features and the rotational angles around the X-axis and Y-axis of the camera.Then,the estimators of regression models,namely the learning features,are used to control the rotational motions around X-axis and Y-axis.The interaction matrix of learning features possess total decoupling and linear properties and has no singularity for any shape of planar objects.Finally,in combination with the normalized centre of gravity features,the normalized area feature and the object orientation feature,a visual servoing controller is designed to conduct the 6-DOF motion control of a camera.Simulated results show that the proposed method is effective.