| The autonomous positioning and track deduction of indoor mobile robots are the basis for autonomous navigation of robots.In indoor environment,the traditional localization method of mobile robot has the disadvantages of low location accuracy,large accumulative error or weak GPS signal.and Combined localization can't solve the fundamental defect of single sensor.The accuracy of visual positioning is not affected by the acceleration change rate of the mobile platform or the side slip of the wheel,only depending on the advantages and disadvantages of the positioning algorithm and the quality of the data collected by the visual sensor.The positioning accuracy of the current vision localization algorithm can meet the requirements of the location under the general conditions,but the amount of computation is huge,which affects the real-time performance of the autonomous positioning of the mobile robot.Therefore,this paper mainly improves the location performance of mobile robot using binocular vision odometer from two aspects of accuracy and real-time.In this paper,the converged binocular positioning measurement model is improved,and the corresponding relationship between the pixel coordinates of the space point in the left and right camera and the physical coordinates in the left camera coordinate system is directly obtained.The improved model is evaluated with the sub-pixel coordinates of the corner point,with an average relative error of 0.10%.Then,the improved ORB algorithm is used to extract the feature points in the image,and the defects of the original ORB algorithm are effectively solved by the uneven distribution of the feature points and the less amount of extraction in the dark light environment.Then,the Flann matching method with higher matching efficiency is used to match the extracted feature points,And the removal of mismatched point pairs is carried out.It is found that in the phase of mismatching between frames,the Homography matrix calculated by the RANSAC algorithm is better than the external point removal of the fundamental matrix.In the frame removal of mismatch,this paper proposes a fundamental matrix F based on two targets as an external point elimination model.Compared with the homography matrix calculated by the RANSAC method,the removal efficiency is almost the same,but the former is only half of the latter.Finally,the track tracking experiment of the binocular vision mileage is done in the static scene in the bright environment.The results show that the BA method,which uses the simultaneous optimization of the position and the map point of the camera,is better than the BA method using only the EPnP method and only the position of the camera.The average track relative error is 0.378%,the mean of the pose error is 1.456%.The FPS of PC under single thread is around 13 frames,which basically meets the location requirement of indoor mobile robot. |
