Research On The Personal Carry Goods Ascription Relation For Robot Personalized Services

In the home environment,when a robot performs a service task,the robot needs to select its exclusive item to perform inference and planning of the task according to the service object that issues the service request instruction,that is,to perform the personalized service.However,current home service robots have a low level of intelligence and insufficient task execution capabilities to meet people's needs for personalized services.How to reason and plan service tasks for specific service objects has become an important issue in improving service levels of service robots.In order to realize personalized service,the service robot is required to know the ascription relation between the people wear articles and the people.Therefore,research on the ascription relation between the people wear articles and the people in the home environment becomes one of the key issues for realizing personalized service in the home environment.The detection,locating,identification of personal carry goods,the identification of robot service objects,and the representation and learning of the ascription relation of objects are problems that need to be solved to realize the robot personalized service.For the sake of providing personalized service for robots,a self-learning framework for ascription relation of personal carry goods is designed.Fusion of deep learning and posture estimation to detect and locate people wear articles.A recognition method of personal carry goods based on convolution neural network is proposed,and combined with the idea of transfer learning,the detection and recognition of personal carry goods is optimized.The identification of human identity in the home environment based on convolution neural network is studied,and the ascription relation memory matrix is formed to realize the binding of the ascription of the personal carry goods based on the vector space cosine similarity selection algorithm.In the home environment,the objects are complicated and the environment is complex.The traditional methods for detection,locating and identification of objects are difficult to achieve good results.Therefore,the article introduces the objects detection and recognition scheme based on the deep learning model,and builds a dataset based on home environment for objects detection and recognition to realize model optimization.The posture estimation model performs the location constraint on the object detection results,and the model can detect and locate the items which are difficult to detect at the same time,thereby effectively improving the objects location ability.Constructing a category-instance two-level objects recognition model,and the principle of image saliency is used to extract the foreground of the objects,achieving a good recognition effect relying only on the category label.The changes of illumination,posture and expression have an important influence on face recognition.How to design a good feature extraction operator to eliminate the influence of these factors on face recognition has been a difficult problem in face recognition field.Therefore,self-constructed multi-light and multi-pose face detection and recognition datasets in home environment is used to train the face detection and recognition models based on the deep learning,and integrate the face posture correction method based on affine transform to further reduce the influence of attitude change to face recognition,and effectively improve the accuracy of face recognition.Based on the detection,positioning and identification of personal carry goods and identification of robot service object,the ascription relation memory matrix is used as the carrier of the ascription relation.The robot obtains multiple ascription relation memory matrices through independent learning of multiple learning cycles.The selection algorithm based on the cosine similarity of vector space completes the selection of the ascription relation memory matrix,eliminates the learned error information and completes the learning of the personal carry goods.The method proposed in this paper closely combines the robots with service needs of humans.The introduction of deep learning has enabled robots to obtain good item identification and service object identification capabilities.The combination with transfer learning has improved the adaptability of robots to the home environment.The autonomous learning of ascription relation provides effective support for robot to provide personalized service.Simulate the home environment in the laboratory to verify the method described in the paper and fully prove the effectiveness of the method which fully proves the effectiveness of the method.This work has important theoretical and practical value for realizing robot personalized service.