Research On 3D Reconstruction Based Robotic Identification And Assembly Methods

Traditional robotic vision relies on 2D color images.Due to its inefficiency and inaccuracy,it is inapplicable to certain scenarios which require high accuracy and high efficiency,such as manipulators in industrial production,home service robots,etc.Especially considering the influence of changing illumination environment conditions and indefinite viewpoint angles,the traditional robotic vision based on the 2D color images cannot even provide a stable output.RGB-D sensor can capture the depth data(point cloud)from the real scenes,which is unaffected by the external factors.Therefore,it is well suited to serve as a robotic vision system.An automated 3D object reconstruction system and an identification and assembly method are proposed in this paper,using the “eye-in-hand” RGB-D sensors.The main research contents of this paper are briefly introduced below:(1)This paper proposes a new view planning method which can be used to calculate the next-best-view(NBV)for multiple robots simultaneously.The entire 3D space is encoded with octree,which marks the voxels with different tags.A set of candidate viewpoints are generated,filtered and evaluated.The viewpoint with highest score is selected as the NBV.And a new algorithm is designed to speed up the generation and filtration of the candidate viewpoints,which can guarantee both speed and quality.(2)An automated 3D object reconstruction system is presented in this paper,which is based on the proposed method and can adapt to various industrial applications.A series of experiments are performed to verify the feasibility of the proposed system.The results show that the proposed system can meet the requirements of industrial production,and has a good performance in the surface coverage and the time cost.(3)A fast 3D object recognition method based on convolutional neural network is proposed in the paper.Considering the practicalities of industrial production,the proposed method makes some modification to the universal model to reach the highest efficiency.(4)A grasp planning method and an automated object grasping system are introduced in this paper.The proposed method takes both the contact area size and the force equilibrium into account.A series of simulations are conducted to demonstrate its practicability and performance.Compared with the existing approaches,it has advantage in the speed and efficiency.