Research Of Object Recognition And Grasping Based On Cloud In Intelligent Space

Intelligent autonomous grasping operation is a great challenge for the research of service robots. At present, the simple intelligence level and insufficient execution ability of service robots can still not live up to the expectations of people. In order to cope with the task of grasping in complex environment, traditional robots focus on improving the capabilities of perception, computing and execution, which makes robots clumsy and expensive. This paper takes object recognition and grasps operation in household environment as application background, based on the combination of intelligent space and cloud, and explores a series of methods to achieve object recognition, pose estimation and grasp algorithm planning in unstructured environment such as home.Focusing on the wide variety of items and complex environment in house, it’s difficult to achieve accurate recognition for objects, apply a method of object recognition and pose estimation based on Vuforia Cloud Service. The Vuforia SDK is used to achieve object recognition and pose estimation through the Vuforia cloud image database. The validity of the method is verified by experiments, and the experimental results are illustrated and analyzed in detail. With the help of the powerful function of Vuforia, the robot can greatly improve its ability of perception, and unload the complex recognition tasks to the cloud, which reduces the burden of storage and computation.The limitations of the recognition service provided by the Vuforia cloud platform are that it’s hard to fully meet the requirements of the recognition task. This paper proposed a method of object classification based on virtual cloud platform of intelligent space. SIFT features and descriptors are detected to build visual vocabulary based on Bag-of-Word model, which realizes the representation of object image. To recognize objects through local cloud platform, a classifier based on SVM algorithm is presented and the accurate object classification of multi-classes is realized. Finally, the experiment is carried out on the local cloud platform, and the validity of the algorithm is verified.Compared with the robot grasping operation in the intelligent space, this paper proposes a combination of platform based on cloud and intelligent space technology. Combined with the bottom robot platform, we build a three-layered architecture with cloud, intelligent space and robot, then design and define the functions of each layer. To analyze the rationality of this architecture, a position-based visual serving (PBVS) control method is applied in the Simulink environment based on the architecture.In order to improve the adaptability of the robot in a cluttered environment, a method based on regression model is proposed to estimate the desired grasping position of the robot. In the period of training data extraction, a method of using six dimensional transformation vectors to represent the pose based on PBVS control law is proposed. Then, a learning training algorithm is designed, and the training data are used to estimate the model. Finally, the parameters of the estimated model are determined by the validation set. The final estimation model is obtained and verified by the test set.