Research On Grasping Object Localization Of Grasping Robot Based On Deep Learning

Grasping robot has a wide range of applications in production.With the development of technology,users have put forward higher requirements for the working efficiency and grasping scope of grasping robot.A good visual system can greatly improve the working efficiency of grasping robot.This thesis firstly locates the target from the two-dimensional image and then obtains the three-dimensional coordinates of the target in the camera coordinate system.Starting from two aspects of two-dimensional positioning and optical three-dimensional positioning of the target,a reliable visual system is researched and designed for accurate threedimensional positioning of the target.Firstly,two-dimensional target location algorithms based on deep learning are studied in this thesis,including YOLOv5 target detection algorithm and Deep Labv3+semantic segmentation algorithm.Lightweight backbone network is used to improve the network due to the large number of Deep Labv3+ parameters and poor real-time performance.In order to solve the problem that the center of the prediction frame of YOLOv5 deviates from the target itself,a segmentation module based on Deep Labv3+ design is added to the YOLOv5 network,and the feature fusion network is designed.The improved network,named YOLOv5-Dep,can achieve instance segmentation of the target,and it outperforms Deep Labv3+ in the segmentation task.Secondly,this thesis studies three-dimensional localization based on binocular vision and binocular structured light.Aiming at the problem of feature matching failure in traditional binocular vision,an optimization Oriented FAST and Rotated BRIEF matching method based on image filtering was proposed.Finally,this thesis uses YOLOv5-Dep network to realize two-dimensional positioning and designs a three-dimensional target positioning system combined with the depth map obtained by the D435 i camera.After the experiment,the robot using this system successfully locates the target,approaches the target according to the obtained target location information,and finally stops 30 cm in front of the target.The experiment shows that the system can play a certain auxiliary role in the grasping process of the grasping robot.