Research On Indoor Vision Localization Technology Based On RGB-D Camera

The positioning techniques represented by inertial navigation systems,ultrasound,WLAN and pseudo-satellite base stations provide technical support for the indoor navigation and location service,but these technologies are either less accurate or expensive and cannot be widely used.With the development of computer vision technology,the optical positioning technology based on camera has become a research hotspot.For computer vision positioning technology,indoor landmarks can provide a large number of spatial attribute information,which can be recognized as prominent and identifiable information in indoor environment.In order to make full use of the existing environmental characteristics,combined with the characteristics of the building itself,proposed a indoor vision positioning technology based on RGB-D camera,the technical proposal draws on the system architecture of visual SLAM,using the RGB-D camera function advantage,design the Pn P and the integration of the ICP motion estimation algorithm,and introduced the closed-loop detection strategy based on key frames mechanism,through the combination of with the method of beam adjustment,provides a more accurate and convenient indoor positioning technology solutions.The main research contents and work are as follows:(1)Introduced the pinhole camera model,analyzed the linear model and nonlinear model of pinhole camera,determines the required for camera calibration parameters;A detailed derivation is carried out on the two-dimensional planar calibration rounds of the camera calibration,established the mathematical model of camera calibration;Elaborated the RGB-D camera depth imaging principle,and the color camera and depth camera of RGB-D camera are calibrated with Open CV respectively.(2)Studied the characteristic detection algorithm of visual odometer,the characteristics of the algorithm features detection principle and the description of the method are analyzed;from the number of feature point extraction accuracy as well as the characteristics of the algorithm,feature point matching the execution speed of the three aspects of feature detection algorithm are compared,and established a research plan based on SURF algorithm;Designed multi-point perspective and iterative closest point method fusion of motion estimation algorithm.The accuracy of motion estimation algorithm is increased by 30%.(3)Described a mathematical model based on graph optimization back-end optimization method,in view of the problems arising from the traditional closed loop testing,this paper proposes a mechanism based on key frames of the closed-loop detection strategy,through to the closed-loop test to establish the global consistency constraints.The overall accuracy of positioning increased by 45%.(4)Built mobile robot hardware and software platform,build the overall framework of the improved visual SLAM system;Introduce that absolute tracking error ATE as the quantitative standard by use the standard dataset to carry out the simulation experiment。 The experiment indicates that the positioning accuracy is relatively high,and ATE is controlled within the decimeter level.Verifying the accuracy and robustness of the proposed visual SLAM system based on RGB-D camera in the indoor environment;Design experimental project of the real site from the linear stability,tested the improved visual SLAM system from a straight line stability,an unclosed loop and a closed loop,verified the validity of the indoor visual localization algorithm based on RGB-D camera.