Research On Loop Closure Detection Based On The Combination Of Point And Line Features

Precise autonomous navigation is one of the key problems that need to be solved when forestry robot carries out precise operation in unknown forest area.Visual SLAM(Simultaneous Localization and Mapping)is the key for forestry robots to realize precise autonomous navigation.Loop closure detection is an important part of visual SLAM,and its task is to identify the places the robot has visited.It can provide correct data correlation for visual SLAM and eliminate the cumulative errors in the calculation process,to improve the accuracy and stability of the SLAM system.At present,most loop closure detection mainly adopts point and line features.In forest environments,the illumination is complex and the color of trees and branches is dim.What's more,the texture of tree branches is similar,which makes the forest environment have serious perception bias problems.The existing methods are inefficient in forest environment,which seriously affects the computational accuracy of visual SLAM systems.Therefore,this paper focus on the loop closure detection in forest environment.In order to make full use of the point and line features in images,a loop closure detection method based on the combination of point and line features is proposed to solve the problem of perception deviation in forest environment.The main contents and contribution in this paper are listed as follows:1.This paper studies the extraction and description algorithms of various features and analyzes their performance in loop closure detection.The BoW model was used to build a visual dictionary based on BREIF SURF ORB LBD binary LBD and apply them in loop closure detection.In most scenarios,the ORB works best in the point feature descriptor and the binary LBD works best in the line feature descriptor.2.An information entropy distribution principle is proposed to calculate image similarity.The information entropy is used to assign weights to the features of points and lines,and the similarity score is obtained.Compared with other principles,the allocation principle proposed in this paper improves the precision of loop closure detection and is more reasonable.3.A loop closure detection algorithm based on the combination of point and line features is proposed.The information entropy distribution principle is used to combine point and line features,and then the similarity score is obtained.Feature matching is introduced to eliminate adjacent images to avoid false positive loop closure.This algorithm improves the precision and recall rate of loop closure detection,extends the application scene of the algorithm,and solves the problem of perception bias in some scene to some extent.4.In this paper,the loop closure detection based on the combination of point and line features is applied to PL-SLAM to realize the positioning and mapping of forestry robot in the forest environment,which provides a theoretical basis for the subsequent automatic navigation technology of forestry robot.