Aiming at the problem of complexity of the scene and occlusion during robot grasping
a object pose estimation method based on depth camera is proposed. Kinect camera is used to capture point cloud image. The FPFH features of point cloud is extracted. Singular value decomposition algorithm and random sample consensus(RANSAC) algorithm are used to estimate the pose of object. The obtained pose is converted into grab posture by hand-eye conversion. Comparing with ICP algorithm and NDT algorithm
the results verify the stability and accuracy of the method.