There is a big error in the traditional method to extract the position and attitude information of the robot. In the process of obtaining the target attitude, a method of extracting the target attitude information of robot arm based on RGB-D data is proposed. The position and attitude of the manipulator target are acquired by depth image processing, and the detected target position is sent to the manipulator control node, and the feature points of the manipulator are extracted. The 3-D mapping is carried out on the acquired RGB image, and the depth and RGB values of feature points, as well as position and attitude information are calculated by using the Gauss mixture model. Finally, the target is extracted by combining the covariance matrix of feature points. The experimental results show that the co-ordinate error and angle error of the robot arm extracted by this method are small. The maximum extraction error is only 28%, which is much lower than the traditional method, which shows that the proposed method is more applicable.