Research On 3D Object Detection And Grasping Technology Based On Deep Learning

In recent years,with the increase in labor costs,it has become a trend to replace manual labor with automation technology.Robots have been widely used in industry,service industry,and medical industry.However,at present,most robots are fixed in a certain position to solve a single task,and traditional manual teaching programming cannot meet the requirements of the new era of "flexibility".Multi-variety,multi-process,small batch and other requirements.Thanks to the improvement of computing power,deep learning has achieved rapid development.How to apply deep learning technology to the field of robots is particularly important,and grasping is an important function of robot sorting and handling operations.Researching the combination of Grasping and deep learning is of great significance for improving the efficiency of robots.In this article,the robotic arm is placed on a mobile chassis combined with a visual sensor as a research object to study its intelligent classification and grasping in different scenarios.In the grasping research,this article studies the expression of 3D object objects,uses deep learning-based 3D object detection to determine the type and position of the object,and uses the detected position for pose calculation,using the results as a guide to grasp.The main contents of this article are as follows:(1)3D model based on the smallest size point.Due to the error between the size of the 3D reconstruction model and the real value,a 3D model method based on the smallest size point is proposed,replacing the complete 3D model with a spatial point,first obtaining the length,width and height information of the object,according to the object length and width High information calculates the point in the object space that can express the 3D object,and finally generates a 3D model with this point.(2)A method for making 3D target detection data set.Since there is a certain gap between the virtually generated data and the real data set,a method for generating and labeling the real data set is proposed.Place the object in the center of the self-made rotating code wheel,move the camera to obtain the data of the object at various angles from different perspectives,and then automatically tag the collected pictures according to the feature points of the object Make.(3)A 3D object detection algorithm based on the smallest size point.The traditional algorithm 3D model is more complicated to produce.On the basis of the existing algorithm,based on the minimum size 3D model of the object,a pose estimation method based on the minimum size model is proposed,and the algorithm is used to target the object Recognition and pose estimation.(4)Mobile robot grasping system.According to the proposed 3D object detection algorithm,the estimated pose is combined with the robot system to construct an intelligent grasping system that can be applied to mobile robots.At the same time,the grasping experiment is verified.Recognize the object and complete the grab.