Object Localization And Management System Based On Multi-Pattern Information Acquisition In Intelligent Space

The existence of a large number of elderly and disabled people has brought many social problems, particularly the urgent requirements for the quantity and quality of the nursing staff. One accredited strategy to deal with this crisis is the development of service robots, which can improve the quality of life for the elders and disabilities, reduce the quantity and quality of the nursing staff, and ensure the stability and development of the society. The basic task of robot is to provide different kinds of services to the elders and disabilities such as taking things, delivering water, and house keeping, etc. In order to accomplish the above tasks, service robots should have the following capabilities:achieve the environmental information accurately, comprehensively and comprehend it in real-time, realize path planning and autonomous navigation, recognize and locate the target in a large scale, and grasp, deliver and place the target in the specified location. Object localization and object management in intelligent space are focused in this paper, and the main contents are listed as follows.(1) A common structure of information representation based on modified JDL and information acquisition based on distributed data fusion tree is presented in the paper according to the features of service robot and intelligent space. In intelligent space, the computing resources and variety of devices are tending to be public and shared, so that the information representation and acquisition mode is different from traditional modes. In order to make full use of the services and the resources provided by heterogeneous sensor network, the multi-pattern information representation model based on modified JDL is built, which can represent multi-class and multi-form information such as QR Code and RFID in an uniform standard, and realizes information intercommunication and sharing. The distributed data fusion tree based multi-pattern information acquisition model is built to provide different information services in data level, feature level and decision level. Furthermore, different fusion algorithms are assigned to appropriate levels. This model can process distributed sensor data in multi-class and multi-level, so that the service robot and intelligent space have the ability to acquire multi-pattern information and can achieve richer information services and sharing.(2) The artificial landmark based distributed environment representation scheme is proposed to overcome the shortcomings of centralized environment representation, such as single storage means, high degree of retrieval difficulty, and low robustness. A three-layer tree hybrid map with topology-topology-grid structure is built to represent indoor environment. The artificial landmark is used to store environmental information, the position and quantity of landmarks are determined by pedigree clustering algorithm according to location of node and distance between node pairs, and the local information in each landmark is optimized using fisheye view. This scheme can reduce the robot’s storage and retrieval load, guarantee the positioning precision, reduce the quantity of landmarks, and improve the efficiency of localization and navigation.A new environment model named danger degree map (DDM) and a modified PSO path planning based on DDM is proposed aiming at overcome the shortcomings of traditional path planning, such as single information representation of grids map and paying inadequate attention to security. The modified PSO algorithm whose fitness function is the weighted sum of the path length and the path danger degree is introduced to get a more reliable and flexible path than traditional method. Non-equidistant distributed PSO method based on the rate of change of barrier is proposed to enhance the adaptability to the environment. The dynamic path planning adopts layered strategy:the modified PSO algorithm is introduced to get a static optimized path and the improved A* algorithm is used to avoid dynamic obstacles according to dynamic DDM. The new methods simplify the planning process, can avoid the obstacle flexibly by making full use of the transcendental environment information and multi-angle information of dynamic obstacle provided by intelligent space, and can meet the real-time requirements of robot navigation in complicated environment.(3) An RFID rough localization system based on Bayes Rule and Particle Filter is proposed to overcome the shortcomings of vision based localization method in large scale. A time sequence based localization method is proposed on the basis of reference tags. It has the features of simple implementation and good real-time performance, but has poor stability. Aiming to improve the stability of localization, the probability model for recognition scope of RFID antenna is built, and then the Bayes Rule and Particle Filter algorithm are introduced to locate the target using the probability model of RFID antenna and the pose of robot. The Bayes Rule based algorithm has the characteristics of high speed, good real-time performance, and high precision. The Particle Filter based algorithm has a higher precision, but a bad real-time performance. The rough localization system makes full use of the advantages of RFID such as wide recognition scope and high search efficiency, which is a beneficial supplement to vision based recognition technology.(4) Aiming at the deficiency that the positioning precision of RFID can’t meet the requirements of object-operating, a multi-type object recognition and accurate-positioning scheme and an object-operating system using position based visual servoing (PBVS) under eye-gaze constraint are presented in the paper. Aiming at the features of objects in home environment such as various types and difficult to recognize, a kind of new artificial object mark based on QR Code is designed to realize fast recognition and get further information of target to provide richer information for object operation. Then an artificial object mark based fast recognition and accurate-positioning scheme for multi-type objects is realized using the system model of robot and the target’s features. Finally, an extended model of 6-DOF manipulator is built, and a PBVS control law under eye-gaze constraint is used to realize the object grasp and delivery. The proposed system utilizes the rich information provided by the robot vision efficiently, and can accurately recognize multi-type objects in complicated indoor environments. Now, the system has been successfully applied to the human-centered active service in intelligent space. (5) Aiming at the low efficiency of object localization and management, an intelligent object management scheme based on Activity Theory is presented. It arrives at the state of object through people’s activity including position and behavior. The RFID tags stuck on the object binding the information space with the physical space, thus the information space can automatically sense the changes in physical space. Object detection, body localization and action recognition are realized using real-time data provided by RFID, distributed vision system and Inertial Navigation Module, combined with database of intelligent space. Object localization oriented Dynamic Bayesian Network is built to automatically sense and update the state of object by the activity analyses of people, which can improve the efficiency of object search and object management greatly.