Research On Vision-Aided Attitude Estimation For Indoor Pedestrian Inertial Navigation

MEMS inertial navigation and positioning technology based on strapdown inertial navigation principle is getting more and more attention.Due to the precision of MEMS gyroscope and the integral solution method of strapdown inertial navigation,the error of carrier attitude estimation accumulates continuously.It's difficult to achieve accurate positioning for a long time.Aiming at application scenario of indoor inertial positioning,the thesis studies the method of using monocular vision to correct the inertial attitude estimation error.The main work is as follows.(1)Visual absolute attitude aids inertial attitude estimation algorithm.The reference object image of indoor characteristic is obtained through visual system,combined with the known reference object attitude.The perspective projection principle is used to obtain absolute angle estimation based on visual information at the current moment.Absolute vision-aided inertial attitude estimation algorithm using adaptive fusion is proposed.Absolute attitude angle of vision and inertial attitude estimation are adaptively weighted to correct the cumulative error of attitude estimation.(2)Visual relative attitude aids inertial attitude estimation algorithm.For continuously collected video,relative rotation angle is estimated by using the principle of epipolar geometry based on difference between adjacent frames.A multi-time scale fusion aids inertial attitude estimation algorithm is proposed,using the characteristics of short-time scale with strong real-time performance and long-time scale with constraining short-time scales.The angular change of different time scales continuously modify and update the inertial attitude.The angle information is corrected multiple times to compensate the offset of gyroscope and the cumulative error of solution,improving the accuracy of attitude angle.(3)Vision-aided inertial attitude estimation fusing algorithm that integrates visual absolute attitude and relative attitude.Fusion of visual absolute attitude aids inertial attitude estimation algorithm and visual relative attitude aids inertial attitude estimation algorithm,the inertial attitude estimation is adaptively weighted with visual absolute attitude angle and multi-timed scale fuses with visual relative rotation angle then.The interaction between static vision and dynamic vision makes full use of line and point features of indoor environment,reducing the error of inertial attitude estimation and improving the accuracy of the carrier attitude estimation.Experiment shows that the combination of visual absolute attitude and visual relative attitude aids the indoor pedestrian inertial attitude estimation algorithm,which can effectively correct the cumulative angle error of the inertial system and improve the accuracy of carrier attitude estimation.During the experimental verification of 4.23 min,the yaw angle error is reduced by 95.33%,the pitch angle error is reduced by 91.67%,and the roll angle error is reduced by 91.21%,the beginning and ending error of the attitude angle can be kept within 4°.