Robot Modeling And Autonomous Positioning Based On Unknown Feature Of Visual Feature

With the developing of economy,society and technology,as well as the increasing trend of social aging,family service robot attract more and more attention by people.The technology of robot localization is key to achieve its own function.However,with the maturing of digital image technology and the advantages of visual information such as large amount of information,easy access and low cost,robot localization technology based on visual information has become the focus by many scholars and experts.In this thesis,aiming at the robot localization in the indoor environment,the ordinary binocular camera as the sensor is used to obtain the visual information,and the extended kalman filter(EKF)is used to realize the robot localization.In order to overcome the shortcomings of traditional simultaneous localization and mapping method,this thesis designed a method of mapping is separate from localization.This method overcomes the problems of traditional EKF algorithm such as weak data association and large amount of computation,improving real-time algorithm.The specific research contents are as follows:Firstly,in this thesis,binocular camera is used as a sensor,binocular camera calibration is very important to remove the camera distortion,which improving accuracy of the process of building plans and localization.Secondly,the progress of establishing a feature-database by an artificial offline algorithm.In this thesis,the SIFT feature is used to describe the environment.Firstly,image library is created.Then the SIFT algorithm is used to extract the features.BBF algorithm based on KD-tree is used to quickly match the left and right images,and the error matching is removed by constraint of binocular vision.In addition,RANSAC is used to improve the correctness of the matching.By using binocular measuring principle,the 3D coordinates of the feature points are restored to complete the 3D reconstruction process.In the 3D reconstruction process,a data associated method is introduced to remove some of the repeated feature points in the feature library,Feature points are too dense,too many feature points will increase the computational complexity and affect the real-time performance.Therefore,the sparse control mechanism of feature points is introduced to control the density and spatial uniformity of feature points in the feature library.At the same time,In order to filter out the high-quality features in the feature library,robust feature point screening strategy was introduced and a feature library based on robust feature points was established,which greatly reduced the number of feature points in the feature library,which lowering the diversity of feature databases.Thirdly,robot localization is based on offline signatures.Firstly,the feature library is matched with the observed features of the current frame image,RANSAC algorithm is used to improve the matching accuracy,which is used to find out the initial position of the robot.After the initial position of the robot is solved,the position of the robot is estimated by using the iterative EKF algorithm iteratively.Finally,in order to improve the environment adaptability of the feature library,in the process of localization,new stable feature points in the environment is updated into the feature library.Finally,study of comprehensive positioning experimental error.The EKF algorithm is used to robot localization in the indoor environment,which is based on each feature library created by the offline process.The database of features in each mode is given separately,and a detailed map of positioning errors is given,and the result of each positioning data is compared.