Research On Dynamic Combined Positioning Method Based On Vision And Inertial Navigation

Autonomous navigation are required for Intelligent robots in the indoor fire rescue,intelligent logistics,warehousing and material transport and so on,and positioning is the basis for autonomous navigation.Visual positioning based on Kinect sensor can obtain rich environmental information and the price is low.However,due to the environment and other factors,the positioning accuracy and the stability is poor.The autonomy of inertial navigation has good autonomy and is free from outside interference and with high accuracy in short term.Therefore,with the help of the correlation and complementarity among multi-source information,the study of visual and inertial navigation for combination positioning aims to improve the stability and reliability of the positioning system.Based on the imaging model of Kinect sensor,a parallel calibration method based on planar template is designed.And the internal and external parameters of the Kinect sensor are estimated.On this basis,the color image and the depth image are registered to realize the fusion of color information and depth information.There are different perspectives and fuzzy in sequence images acquired by mobile Kinect,which leads to the problem of uncertainty of the feature matching between frames.Based on the number of extracted feature points and the matching number,three feature extraction algorithms which is commonly used are analyzed for invariance of image rotation and degraded quality.Through the experimental comparison,the SURF feature extraction matching algorithm with good stability is determined.The mis-matching phenomenon of feature points was analyzed,and the algorithm of two dimensional feature points mis-matching elimination was improved,which improves theaccuracy of feature matching.The ICP three-dimensional point set registration algorithm that integrates depth information is designed,and the rotation matrix and translation vector are estimated to realize motion estimation and visual positioning.Aiming at the problem that the localization performance of the Kinect sensor is affected by the environment and its own factors,the method of dynamic integrated positioning based on the fusion inertial navigation is studied.Temporal and spacial alignment of the data of inertial navigation and Kinect sensors are carried out.The number of the correct matching feature points and depth error are proposed as the judging condition of the combination strategy,and a fusion algorithm based on extended Calman filtering combined positioning is designed.We can use the positioning data of the inertial navigation to construct a process model and to fuse the position and pose estimation results of the Kinect sensor visual positioning.Furthermore,the location of the Kinect sensor is selectively modified.The experiment of robot positioning was carried out in indoor laboratory and aisle corridor.The results show that INS and visual combination positioning method can improve the positioning stability and reliability effctively.