Visual Closed-loop Detection Based Features Of Mobile Robot

How to solve the simultaneous localization and mapping(SLAM)under an unknown environment has always been an unavoidable key technical problem for robots.As a basic step in the SLAM framework,Closed-loop detection plays an important role in reducing the cumulative error of the robot's front end and map optimization on the back end.Closed-loop detection technology refers to comparing the similarity between the current image frames and the historical image frames which obtained by the mobile robot to determine whether there is a closed loop in the robot's route,and then synchronously update the environment map based on the determination result.Now the visual closed-loop algorithm is the mainstream detection.Although the research results of visual closed-loop are numerous,there are still some deficiencies.Therefore,this paper focus on the problems which exists in the visual closed-loop detection algorithm.The main contents are as follows:Firstly,this paper outlines the research background and significance of the thesis topic,introduces the basic theory of SLAM for robot vision,analyzes the mainstream algorithms of visual closed-loop detection and its key issues in detail which provides the theoretical model and basis for subsequent research.Secondly,the paper proposes the multi-scale ORB+FREAK feature detection algorithm for the closed-loop detection of visual robots which overcome the shortcomings,time-consuming and memory-intensive of SURF feature algorithm detection.For each image,a corresponding scale space is constructed,and then the ORB algorithm is used to detect the feature points in the scale space,and the feature description uses the FREAK descriptors.And the simulation experiments prove that the improved algorithm not only improves the detection speed but also reduces the memory loss.Thirdly,the disadvantage of the K-means algorithm in the traditional BOVW model is sensitive to the initial clustering center value and is easy to fall into the local optimal value,so this paper combines the K-means clustering algorithm with the improved artificial bee colony to improve the clustering accuracy,and then use this new algorithm tobuild layered BoVW model.Also through the experimental results,the reliability of the algorithm was verified by the image classification experiments.Finally,for the problem that the accuracy of the closed-loop detection algorithm only uses the image matching method of local feature points is not high,so the global features and the tracking model are introduced.Based on the content of the first two chapters,the feature extraction and feature description of the image are performed.and build the improved BoVW dictionary.First use the global feature and the tracking model detect the initial closed-loop,then calculate the similarity score of each image,use the similarity of the pyramid image to match each score,and assure the final closed-loop.Through closed-loop experiments of indoor and outdoor image sequences,the effectiveness of the proposed algorithm is verified.