Aiming at the intelligent application of industrial manipulators, the main research is based on machine vision-based intelligent gripping of manipulators. First, noise reduction is performed based on the point cloud data, and then the grasping pose data set is obtained through the sampling algorithm and combined with the CNN network to achieve pose estimation without pre-building object model. Finally, by building a robotic arm control platform based on ROS, the robotic arm grasping based on the deep learning method and the grasping based on the traditional grasping method are completed, and the two methods are analyzed. The experimental results show that the grasping posture acquired by the deep learning method is more conducive to grasping than the traditional grasping method.