Font Size: a A A

Monocular SLAM With IMU Information

Posted on:2021-05-19Degree:MasterType:Thesis
Country:ChinaCandidate:C J DongFull Text:PDF
GTID:2428330611471503Subject:Vehicle engineering
Abstract/Summary:PDF Full Text Request
SLAM technology is to solve the local positioning of the carrier(such as a vehicle)without prior information.The development of visual SLAM is particularly important for the application of SLAM technology in automatic driving positioning.This article is based on the ORB-SLAM system.According to the characteristics of IMU,the inertial-visual tracking method is added to the tracking thread,and the optimization items of IMU related variables are added to the optimization cost function.In the local optimization thread,the IMU measurement must be taken into account It makes sense only under continuous conditions.Some improvements have been made to the structure of the optimization problem.The specific research contents are as follows:(1)Using the method of mutual information,the feature points of ORB-SLAM are screened to reduce the number of map points.First of all,in this paper,we use the experimental method to determine the approximate range of the mutual information of the feature points in the image.After obtaining the approximate range,this paper uses different mutual information thresholds to filter the feature points in the ORB-SLAM system to obtain a reasonable threshold.(2)IMU measurement is added on the basis of visual SLAM,which improves the positioning robustness and outdoor long-term positioning accuracy.In the tracking thread,a visual-inertial tracking method is added to replace the original visual tracking method.In the local optimization thread,the process of IMU initialization is added,and the original visual optimization method is replaced by a visual-inertial optimization method.(3)Adjust some strategies in ORB-SLAM to make it more suitable for SLAM system after fusing IMU information.After the IMU measurement is added,the key frame insertion method of visual tracking and the optimization problem construction method are no longer applicable.This paper makes corresponding modifications to the sensor characteristics of the IMU.(4)On the KITTI and EUROC data sets,the text SLAM system is compared with Vins-Mono,ORB-SLAM,which proves that the improved SLAM system has higher accuracy.Experiments show that the SLAM system in this paper has high accuracy both indoors and outdoors,while ORB-SLAM can only guarantee the accuracy indoors.
Keywords/Search Tags:Autonomous driving, visual odometry, visual-inertial SLAM, mutual information, pre-integration technology
PDF Full Text Request
Related items