Research On Intelligent Vehicle Monocular Visual Odometer Algorithm

Monocular vision odometer is widely used in mobile robots and intelligent vehicles because of its low memory consumption,simple configuration and low price.However,traditional monocular vision odometer based on feature point matching method also has some problems,such as poor real-time performance,scale ambiguity and large cumulative error.For this reason,this paper carries out some algorithm optimization research on monocular vision odometer.In order to improve the real-time performance of monocular vision mileage calculation method,SURF(Speeded Up Robust Feature)algorithm and ORB(Oriented Fast and Rotated Brief)algorithm are studied.An improved ORB is proposed based on Hessin matrix feature detection and ORB r BRIEF descriptor,aiming at the problem of poor real-time performance of SURF algorithm and scale invariance of ORB algorithm.In order to improve the scale ambiguity and cumulative error of monocular vision mileage calculation method,the optical flow motion estimation algorithm is analyzed.The problem of ground tracking loss in EEM(Edge Expand Method)is optimized.The scale ambiguity problem of monocular vision mileage calculation method is solved,and the continuous tracking of ground features is realized.Based on KALMAN filter,a monocular vision odometer fusion algorithm is designed.Optical flow method is used to obtain more position feature information,which improves the accuracy of vehicle motion estimation and reduces the cumulative error.Based on KITTI open data set,vehicle simulation experiments are carried out.The trajectory,rotation and translation error characteristics of the flow method,feature point matching method and their fusion algorithm are analyzed.The real-time performance,scale ambiguity and cumulative error of the monocular vision mileage calculation method are improved,and the feasibility of the proposed monocular vision odometer fusion algorithm is verified.