Research On Scale Acquisition And Locating Method Of Monocular Vision Inertial Odometer

With the rise of computer vision,the vision odometer,which can capture the motion information of the carrier by taking the image of the surrounding environment with the vision sensor,has become a new development direction of navigation and positioning technology.However,the image quality of the visual sensor is easily affected by environment,so it is not usually used alone.Inertial devices can provide stable carrier motion data without environmental interference and can complement the advantages of the visual navigation devices,so visual inertial odometer becomes an important research direction in the field of navigation and positioning.Visual inertial odometer can be divided into monocular,binocular and multi-ocular vision odometer from the perspective of vision,and only binocular or monocular vision odometer are usually used in cost-considering solutions.Binocular cameras capture more information than monocular cameras,so the settlement process is simpler.However,when the distance between the binocular camera and the subject is much longer than the baseline length of the binocular(i.e.the distance between the optical centers of the two cameras),binocular vision will degenerate into monocular vision,and the binocular vision requires the two cameras to be identical,as a result,the cost of binocular vision is much higher than monocular vision.Therefore,it is of great significance to study monocular vision odometer.In the monocular vision inertia odometer,it is difficult to obtain the depth of field information of the surrounding environment only when the images are taken by monocular camera,which leads to the uncertainty of the scale of the carrier displacement information calculated by monocular vision,and thus the accurate value of the carrier motion cannot be obtained,only the trajectory with normalized scale can be obtained.Aiming at this problem,this paper proposes a visual scale access method based on inertia calculating.Firstly,the visual scale acquisition method under the certain initial displacement is realized,and based on this method,the precision of Inertial Navigation System’s initial displacement is analyzed.And thus a new method of getting the visual scale is presented,which is based on the initial displacement of the inertia calculating.Finally,an improvement is made on the basis of this method.An inertial auxiliary visual scale acquisition method based on motion state decision is implemented,and the feasibility of the above method is verified by experiments.Monocular vision inertial odometers combine the advantages that the precision of vision is still considerable and the inertia is not easily disturbed by the environment when the vision is working for a long time.In this paper,the Kalman filtering method is adopted to realize the fusion of monocular vision and inertial solution information.However,the precision of simple monocular vision inertial data fusion is limited in the practical application.In this paper,a visual inertial coordinate alignment algorithm based on hand-eye calibration and a visual data processing optimization Algorithm based on Histogram equalization are proposed to improve the overall accuracy of monocular visual inertial odometers.Finally,the feasibility of the optimization method and the applicability of the monocular odometry in different application scenarios are verified by the actual experimental results.