Closed-loop Detection Based Vision SLAM Of Mobile Robot

The closed loop detection of mobile robot Simultaneous Localization And Mapping(SLAM)is an important part of the robot to judge whether it needs to update the map.Due to the error caused by the robot's vision sensor,the accuracy of the closed loop detection cannot be guaranteed.Therefore,in this paper,the existing problems of the important steps in the closed-loop detection algorithm are studied.The major research contents are as follows:First,the significance and background of the research are introduced.The research status and some key problems of the closed-loop detection algorithm are analyzed.And the basic methods of SLAM and some characteristic detection algorithms in closed-loop detection are analyzed in detail.Which providing the theoretical model and basis is for the following work.Second,in view of the problems of slow image processing speed and accuracy of feature points matching in mobile robot vision-based SLAM.The paper proposes a novel matching method based on color feature and improved Speeded Up Robust Features(SURF)algorithm.First,image sequences are roughly matched by color characteristics,and make sure five images that is most similar to the test image.Then,Hessian matrix is used to get the feature points that described by Krawtchouk moments,and calculate the feature points' gradient direction and amplitude.And then,the new feature vector has been obtained.The simulation experiment proves that proposed algorithm improves the precision and efficiency of image matching.Again,an improved Bag Of Visual Words(BoVW)method is designed for the low accuracy of the BoVW model in closed-loop detection.The SURF operator based on Krautchouk moments is used to extract the features of the de-fogged scene based on the guidance filter.Using the maximum and minimum distance function combining particle swarm clustering algorithm to construct visual words,the contrastive experiments of image classification are carried out to prove that the proposed algorithm improves the accuracy rate.Finally,for the problem of the lack of accuracy in closed-loop detection method,the new closed-loop detection method based on posterior processing is designed.The SURF algorithm based on Krautchouk moment is used to extract the image features,and a pyramid hierarchical visual word tree is constructed.And the similarity measure is calculated by the modified weighted Term Frequency-Inverse Document Frequency(TF-IDF)entropy score calculation method.The initial closed loop is determined by comparing with the threshold value,and the correct closed loop is obtained by posterior processing of time constraint and geometric relation of polar line.The experiments are carried out to verify the feasibility of the proposed algorithm.