Research On Slam Closed-loop Detection For Indoor Scenes

The technology required in the continuous development of mobile robots is being updated and improved in real time.The research on the visual SLAM technology required for robot positioning is also constantly deepened,and the closed-loop detection link in visual SLAM is crucial to the accuracy of overall positioning.However,the current closed-loop detection method still has some problems,so to improve the problems in closed-loop detection,a SLAM closed-loop detection research for indoor scenes is proposed.The main research contents and innovations are as follows:(1)The detection method,description method and matching method of image feature points are introduced in detail.First of all,for the detection method of image feature points,the two typical SIFT and FAST algorithms are introduced in detail from the aspects of algorithm implementation steps and algorithm characteristics.Secondly,for the description method of image feature points,it mainly introduces the gradient-based histogram method and the binary-based method,and analyzes the differences and advantages and disadvantages of the two in describing the speed.Thirdly,on this basis,the extraction method and matching method of image feature points are improved,and the SIFT algorithm is used in feature point detection;rBRIEF is used to describe the detected key points in feature point description;feature point matching At the same time,first use the spatial consistency principle to remove part of the feature points,remove some of the feature points that are similar but not identical,and then use the random sampling consistency algorithm to match the image feature points to complete the scene image matching process.(2)According to the detailed analysis of the existing closed-loop detection methods,and the problems of perceptual ambiguity and low accuracy of the common closed-loop detection methods,an improved method using hierarchical visual dictionary trees and TF-IDF entropy score matching is proposed.The detection method is improved to solve the existing problems.First,in the improved algorithm,the flat structure of the visual dictionary tree is replaced by a hierarchical visual dictionary tree,so that not only can you obtain more visual words,but also reduce some of the computational complexity;second,there is about the similarity of scene images In the calculation part of,the improved method of TF-IDF entropy score matching is used.Finally,in order to ensure that the correct closed loop is detected,the posterior processing link is added,that is,the constraints of the robot's running time and running space are used to eliminate the error closed loop that does not meet the temporal and spatial constraints,thereby increasing the accuracy of detecting the closed loop degree.(3)Introduce the construction of the experimental platform and specific closed-loop detection experiments.First,in order to verify the difference between the ORB algorithm and the improved algorithm in this paper,perform image feature point extraction and matching experiments,and analyze the performance of the improved algorithm;then,the improved method of TF-IDF entropy score matching is applied to closed-loop detection experiments Based on the comparison of closed-loop detection methods based on SIFT algorithm,the experimental results verify that the improved closed-loop detection method has certain advantages in terms of accuracy.In order to better verify the algorithm in this paper,some experimental data are selected from the TUM data set for verification experiments,and the selected standard data set is used to improve the closed-loop detection method for experimentation,and then the root mean square error(ATE RMSE)is used as the evaluation of closed-loop detection The method is compared with the traditional RGB-D SLAM method.It is verified that the accuracy of the absolute trajectory error of the improved method is better than that of the traditional RGB-D SLAM system.